diff --git a/README.md b/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..1d5830ba265bc6986c8c2560ed47a846bd6a09e4
--- /dev/null
+++ b/README.md
@@ -0,0 +1,147 @@
+---
+tags:
+- generated_from_trainer
+model-index:
+- name: zephyr-7b-alpha
+ results: []
+license: mit
+datasets:
+- stingning/ultrachat
+- openbmb/UltraFeedback
+language:
+- en
+base_model: mistralai/Mistral-7B-v0.1
+---
+
+
+
+
+
+
+# Model Card for Zephyr 7B Alpha
+
+Zephyr is a series of language models that are trained to act as helpful assistants. Zephyr-7B-α is the first model in the series, and is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) that was trained on on a mix of publicly available, synthetic datasets using [Direct Preference Optimization (DPO)](https://arxiv.org/abs/2305.18290). We found that removing the in-built alignment of these datasets boosted performance on [MT Bench](https://huggingface.co/spaces/lmsys/mt-bench) and made the model more helpful. However, this means that model is likely to generate problematic text when prompted to do so.
+
+
+## Model description
+
+- **Model type:** A 7B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
+- **Language(s) (NLP):** Primarily English
+- **License:** MIT
+- **Finetuned from model:** [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
+
+### Model Sources
+
+
+
+- **Repository:** https://github.com/huggingface/alignment-handbook
+- **Demo:** https://huggingface.co/spaces/HuggingFaceH4/zephyr-chat
+
+## Intended uses & limitations
+
+The model was initially fine-tuned on a variant of the [`UltraChat`](https://huggingface.co/datasets/stingning/ultrachat) dataset, which contains a diverse range of synthetic dialogues generated by ChatGPT. We then further aligned the model with [🤗 TRL's](https://github.com/huggingface/trl) `DPOTrainer` on the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset, which contain 64k prompts and model completions that are ranked by GPT-4. As a result, the model can be used for chat and you can check out our [demo](https://huggingface.co/spaces/HuggingFaceH4/zephyr-chat) to test its capabilities.
+
+Here's how you can run the model using the `pipeline()` function from 🤗 Transformers:
+
+```python
+# Install transformers from source - only needed for versions <= v4.34
+# pip install git+https://github.com/huggingface/transformers.git
+# pip install accelerate
+
+import torch
+from transformers import pipeline
+
+pipe = pipeline("text-generation", model="HuggingFaceH4/zephyr-7b-alpha", torch_dtype=torch.bfloat16, device_map="auto")
+
+# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating
+messages = [
+ {
+ "role": "system",
+ "content": "You are a friendly chatbot who always responds in the style of a pirate",
+ },
+ {"role": "user", "content": "How many helicopters can a human eat in one sitting?"},
+]
+prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
+print(outputs[0]["generated_text"])
+# <|system|>
+# You are a friendly chatbot who always responds in the style of a pirate.
+# <|user|>
+# How many helicopters can a human eat in one sitting?
+# <|assistant|>
+# Ah, me hearty matey! But yer question be a puzzler! A human cannot eat a helicopter in one sitting, as helicopters are not edible. They be made of metal, plastic, and other materials, not food!
+```
+
+## Bias, Risks, and Limitations
+
+
+
+Zephyr-7B-α has not been aligned to human preferences with techniques like RLHF or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so).
+It is also unknown what the size and composition of the corpus was used to train the base model (`mistralai/Mistral-7B-v0.1`), however it is likely to have included a mix of Web data and technical sources like books and code. See the [Falcon 180B model card](https://huggingface.co/tiiuae/falcon-180B#training-data) for an example of this.
+
+
+## Training and evaluation data
+
+Zephyr 7B Alpha achieves the following results on the evaluation set:
+
+- Loss: 0.4605
+- Rewards/chosen: -0.5053
+- Rewards/rejected: -1.8752
+- Rewards/accuracies: 0.7812
+- Rewards/margins: 1.3699
+- Logps/rejected: -327.4286
+- Logps/chosen: -297.1040
+- Logits/rejected: -2.7153
+- Logits/chosen: -2.7447
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+
+- learning_rate: 5e-07
+- train_batch_size: 2
+- eval_batch_size: 4
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 16
+- total_train_batch_size: 32
+- total_eval_batch_size: 64
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- lr_scheduler_warmup_ratio: 0.1
+- num_epochs: 1
+
+### Training results
+
+| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 0.5602 | 0.05 | 100 | 0.5589 | -0.3359 | -0.8168 | 0.7188 | 0.4809 | -306.2607 | -293.7161 | -2.6554 | -2.6797 |
+| 0.4852 | 0.1 | 200 | 0.5136 | -0.5310 | -1.4994 | 0.8125 | 0.9684 | -319.9124 | -297.6181 | -2.5762 | -2.5957 |
+| 0.5212 | 0.15 | 300 | 0.5168 | -0.1686 | -1.1760 | 0.7812 | 1.0074 | -313.4444 | -290.3699 | -2.6865 | -2.7125 |
+| 0.5496 | 0.21 | 400 | 0.4835 | -0.1617 | -1.7170 | 0.8281 | 1.5552 | -324.2635 | -290.2326 | -2.7947 | -2.8218 |
+| 0.5209 | 0.26 | 500 | 0.5054 | -0.4778 | -1.6604 | 0.7344 | 1.1826 | -323.1325 | -296.5546 | -2.8388 | -2.8667 |
+| 0.4617 | 0.31 | 600 | 0.4910 | -0.3738 | -1.5180 | 0.7656 | 1.1442 | -320.2848 | -294.4741 | -2.8234 | -2.8521 |
+| 0.4452 | 0.36 | 700 | 0.4838 | -0.4591 | -1.6576 | 0.7031 | 1.1986 | -323.0770 | -296.1796 | -2.7401 | -2.7653 |
+| 0.4674 | 0.41 | 800 | 0.5077 | -0.5692 | -1.8659 | 0.7656 | 1.2967 | -327.2416 | -298.3818 | -2.6740 | -2.6945 |
+| 0.4656 | 0.46 | 900 | 0.4927 | -0.5279 | -1.6614 | 0.7656 | 1.1335 | -323.1518 | -297.5553 | -2.7817 | -2.8015 |
+| 0.4102 | 0.52 | 1000 | 0.4772 | -0.5767 | -2.0667 | 0.7656 | 1.4900 | -331.2578 | -298.5311 | -2.7160 | -2.7455 |
+| 0.4663 | 0.57 | 1100 | 0.4740 | -0.8038 | -2.1018 | 0.7656 | 1.2980 | -331.9604 | -303.0741 | -2.6994 | -2.7257 |
+| 0.4737 | 0.62 | 1200 | 0.4716 | -0.3783 | -1.7015 | 0.7969 | 1.3232 | -323.9545 | -294.5634 | -2.6842 | -2.7135 |
+| 0.4259 | 0.67 | 1300 | 0.4866 | -0.6239 | -1.9703 | 0.7812 | 1.3464 | -329.3312 | -299.4761 | -2.7046 | -2.7356 |
+| 0.4935 | 0.72 | 1400 | 0.4747 | -0.5626 | -1.7600 | 0.7812 | 1.1974 | -325.1243 | -298.2491 | -2.7153 | -2.7444 |
+| 0.4211 | 0.77 | 1500 | 0.4645 | -0.6099 | -1.9993 | 0.7656 | 1.3894 | -329.9109 | -299.1959 | -2.6944 | -2.7236 |
+| 0.4931 | 0.83 | 1600 | 0.4684 | -0.6798 | -2.1082 | 0.7656 | 1.4285 | -332.0890 | -300.5934 | -2.7006 | -2.7305 |
+| 0.5029 | 0.88 | 1700 | 0.4595 | -0.5063 | -1.8951 | 0.7812 | 1.3889 | -327.8267 | -297.1233 | -2.7108 | -2.7403 |
+| 0.4965 | 0.93 | 1800 | 0.4613 | -0.5561 | -1.9079 | 0.7812 | 1.3518 | -328.0831 | -298.1203 | -2.7226 | -2.7523 |
+| 0.4337 | 0.98 | 1900 | 0.4608 | -0.5066 | -1.8718 | 0.7656 | 1.3652 | -327.3599 | -297.1296 | -2.7175 | -2.7469 |
+
+
+### Framework versions
+
+- Transformers 4.34.0
+- Pytorch 2.0.1+cu118
+- Datasets 2.12.0
+- Tokenizers 0.14.0
\ No newline at end of file
diff --git a/added_tokens.json b/added_tokens.json
new file mode 100644
index 0000000000000000000000000000000000000000..cbce74e5c64b97114098962fa58454a57d7fb532
--- /dev/null
+++ b/added_tokens.json
@@ -0,0 +1,5 @@
+{
+ "": 2,
+ "": 1,
+ "": 0
+}
diff --git a/all_results.json b/all_results.json
new file mode 100644
index 0000000000000000000000000000000000000000..a32a5840be00cf99ad0d5ecb2da8313326c56a73
--- /dev/null
+++ b/all_results.json
@@ -0,0 +1,21 @@
+{
+ "epoch": 1.0,
+ "eval_logits/chosen": -2.744652509689331,
+ "eval_logits/rejected": -2.71529483795166,
+ "eval_logps/chosen": -297.10400390625,
+ "eval_logps/rejected": -327.4286193847656,
+ "eval_loss": 0.46045970916748047,
+ "eval_rewards/accuracies": 0.78125,
+ "eval_rewards/chosen": -0.5052940249443054,
+ "eval_rewards/margins": 1.3699172735214233,
+ "eval_rewards/rejected": -1.8752113580703735,
+ "eval_runtime": 52.3218,
+ "eval_samples": 1000,
+ "eval_samples_per_second": 19.113,
+ "eval_steps_per_second": 0.306,
+ "train_loss": 0.48803097129668166,
+ "train_runtime": 7971.1784,
+ "train_samples": 61966,
+ "train_samples_per_second": 7.774,
+ "train_steps_per_second": 0.243
+}
\ No newline at end of file
diff --git a/cal_data.safetensors b/cal_data.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cd988e09571b1e2a570a608f858c02db1d47b325
--- /dev/null
+++ b/cal_data.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08be1103ff8fcef33b570f3c0f5ae4cc7f9dc5c3f264105baa55fc9b132ed1be
+size 1638488
diff --git a/colab-demo.ipynb b/colab-demo.ipynb
new file mode 100644
index 0000000000000000000000000000000000000000..ca05601599bcac111a68107ee578e9df4add4fff
--- /dev/null
+++ b/colab-demo.ipynb
@@ -0,0 +1 @@
+{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"provenance":[],"gpuType":"T4","authorship_tag":"ABX9TyNcXbnhVLPdImfXNrkyZZK9"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"},"accelerator":"GPU","widgets":{"application/vnd.jupyter.widget-state+json":{"af35743bd7cc4b2a9a9ac89166372f1f":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_cc8496246b9248ae8631168804de7374","IPY_MODEL_0a9a7266561e40a991473c40721d1caa","IPY_MODEL_83bb42e8a79446cfa6bdfb76284a2f2c"],"layout":"IPY_MODEL_a3df2076447a4f5e8b777295f55c8fb4"}},"cc8496246b9248ae8631168804de7374":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_cb54c0578558402bb0b8341999449b92","placeholder":"","style":"IPY_MODEL_a76ea86b1887442e8828d1184f548b05","value":"Downloading (…)lve/main/config.json: 100%"}},"0a9a7266561e40a991473c40721d1caa":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_9659515b9831421d9464ffee29b2b87a","max":628,"min":0,"orientation":"horizontal","style":"IPY_MODEL_a08f5e40fb1e4662aa624e6feb3eddac","value":628}},"83bb42e8a79446cfa6bdfb76284a2f2c":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_790926668e1e4c4ab390c0821529c3c1","placeholder":"","style":"IPY_MODEL_d3b27018eaf34defa08ff8c0c09ef826","value":" 628/628 [00:00<00:00, 44.8kB/s]"}},"a3df2076447a4f5e8b777295f55c8fb4":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"cb54c0578558402bb0b8341999449b92":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"a76ea86b1887442e8828d1184f548b05":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"9659515b9831421d9464ffee29b2b87a":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"a08f5e40fb1e4662aa624e6feb3eddac":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"790926668e1e4c4ab390c0821529c3c1":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"d3b27018eaf34defa08ff8c0c09ef826":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"2f7c528ac4a846b3bd3f5b859c5937ce":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_779303ae3a504bc39e94c05bb41d6680","IPY_MODEL_12b3162cbf9c4c3ba4f83f325341fc26","IPY_MODEL_ed0a166c01104946b3768f827f9a92da"],"layout":"IPY_MODEL_99f491f558e248abb4c7acc9deb28923"}},"779303ae3a504bc39e94c05bb41d6680":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_91eaa2a5525f4f51a111ad59dca862dd","placeholder":"","style":"IPY_MODEL_70c24596219147189534ef2effd75212","value":"Downloading (…)fetensors.index.json: 100%"}},"12b3162cbf9c4c3ba4f83f325341fc26":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_e54d337cbae84671ba46d2fccf7241a5","max":23950,"min":0,"orientation":"horizontal","style":"IPY_MODEL_586b2f72d4694e3c88f22131326378b6","value":23950}},"ed0a166c01104946b3768f827f9a92da":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_3abcf30b8b7a49d480d80b43f7528f3b","placeholder":"","style":"IPY_MODEL_030709c0956f400c9c61d9c5028a7b5a","value":" 23.9k/23.9k [00:00<00:00, 526kB/s]"}},"99f491f558e248abb4c7acc9deb28923":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"91eaa2a5525f4f51a111ad59dca862dd":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"70c24596219147189534ef2effd75212":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"e54d337cbae84671ba46d2fccf7241a5":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"586b2f72d4694e3c88f22131326378b6":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"3abcf30b8b7a49d480d80b43f7528f3b":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"030709c0956f400c9c61d9c5028a7b5a":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"c2e47ceae1d14da59867cc5653409e7e":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_07b3f2f9ab8e479489cfa96d8ced3f4c","IPY_MODEL_5bd6352bb7bf4c2fa195c73d9b875ea7","IPY_MODEL_02eab2611e47411eb957e42321093df2"],"layout":"IPY_MODEL_4dec5009208844fc845d276f86c53ea2"}},"07b3f2f9ab8e479489cfa96d8ced3f4c":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_cdde97c62f854f8f96a86957c4e67487","placeholder":"","style":"IPY_MODEL_1fa0cf1349ed4f779a3f61f8f5be0f1e","value":"Downloading shards: 100%"}},"5bd6352bb7bf4c2fa195c73d9b875ea7":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_f28f325a3aa24182856e5da82b565ddd","max":8,"min":0,"orientation":"horizontal","style":"IPY_MODEL_56e4af5782d54e9c8069444b34fc6b79","value":8}},"02eab2611e47411eb957e42321093df2":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_bbad101d94d64d3e8f47f2d22bf3096e","placeholder":"","style":"IPY_MODEL_33f0fd3841ac42788e92041f2756df4a","value":" 8/8 [02:06<00:00, 13.41s/it]"}},"4dec5009208844fc845d276f86c53ea2":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"cdde97c62f854f8f96a86957c4e67487":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"1fa0cf1349ed4f779a3f61f8f5be0f1e":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"f28f325a3aa24182856e5da82b565ddd":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"56e4af5782d54e9c8069444b34fc6b79":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"bbad101d94d64d3e8f47f2d22bf3096e":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"33f0fd3841ac42788e92041f2756df4a":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"6cf085e28dc444698da1fcb0671d5570":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_1d2644bfdfa1493bb682f9bf80e132be","IPY_MODEL_d28ec950aa0d4c5b817e54172da5fb44","IPY_MODEL_db47d15652ac4a51b85cbb0e1f49cffc"],"layout":"IPY_MODEL_6148302fcdcd4eeea8770340bcdc4af5"}},"1d2644bfdfa1493bb682f9bf80e132be":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_09c91af5582f40d1b864a3f1a6728190","placeholder":"","style":"IPY_MODEL_a3cc8c7cdec74aa3a42ae888fd12b813","value":"Downloading (…)of-00008.safetensors: 100%"}},"d28ec950aa0d4c5b817e54172da5fb44":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_0977d1f5e8f54ee1ae13509423b6a530","max":1889587040,"min":0,"orientation":"horizontal","style":"IPY_MODEL_5695943f70204d65aac3d6a994fcac9c","value":1889587040}},"db47d15652ac4a51b85cbb0e1f49cffc":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_8a2f7e2ba8254dd8929538d88c1be72e","placeholder":"","style":"IPY_MODEL_e5981daa2da24c299dec59442d3cd3d9","value":" 1.89G/1.89G [00:18<00:00, 26.5MB/s]"}},"6148302fcdcd4eeea8770340bcdc4af5":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"09c91af5582f40d1b864a3f1a6728190":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"a3cc8c7cdec74aa3a42ae888fd12b813":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"0977d1f5e8f54ee1ae13509423b6a530":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"5695943f70204d65aac3d6a994fcac9c":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"8a2f7e2ba8254dd8929538d88c1be72e":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"e5981daa2da24c299dec59442d3cd3d9":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"5660194f98bd47f9a872fb3ba194e722":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_287f1b8059ff461580964d6765505048","IPY_MODEL_6106c56288dc46fe812176811ede8154","IPY_MODEL_faf9c3ef160d4772b6b4d74b928f8df8"],"layout":"IPY_MODEL_ba143a50545a4c5588ffa381b3d86457"}},"287f1b8059ff461580964d6765505048":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_9d06d6845ae84d3ebfc61d1414c141ab","placeholder":"","style":"IPY_MODEL_baae3ea716dc4c26bc1720f25b2eee60","value":"Downloading (…)of-00008.safetensors: 100%"}},"6106c56288dc46fe812176811ede8154":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_c667c1db826741f0ae7e8954071b953e","max":1946243936,"min":0,"orientation":"horizontal","style":"IPY_MODEL_89f63cd121554cbbb201cb999ac17515","value":1946243936}},"faf9c3ef160d4772b6b4d74b928f8df8":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_6e0d6d6061924a9080e46924e0a185d5","placeholder":"","style":"IPY_MODEL_4a73e20d2402485fa864b16e0cfe40e8","value":" 1.95G/1.95G [00:19<00:00, 119MB/s]"}},"ba143a50545a4c5588ffa381b3d86457":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"9d06d6845ae84d3ebfc61d1414c141ab":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"baae3ea716dc4c26bc1720f25b2eee60":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"c667c1db826741f0ae7e8954071b953e":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"89f63cd121554cbbb201cb999ac17515":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"6e0d6d6061924a9080e46924e0a185d5":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"4a73e20d2402485fa864b16e0cfe40e8":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"ebf0e040eeee4b2896f03f4c6eb2b297":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_b235c4229aa943759638c05f7ee365ce","IPY_MODEL_111c3f38704d432b828946a63c909381","IPY_MODEL_3671af40a3944553bff108497c34d656"],"layout":"IPY_MODEL_ede441a2ff5a4970a9db369487dd91a6"}},"b235c4229aa943759638c05f7ee365ce":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_92de8ec1a92a487eb872698185359a27","placeholder":"","style":"IPY_MODEL_56dcdf86a9ad4342ac63a1d7c19bdbbd","value":"Downloading (…)of-00008.safetensors: 100%"}},"111c3f38704d432b828946a63c909381":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_fb49971dab3749bd854b68572fea4eb9","max":1979781432,"min":0,"orientation":"horizontal","style":"IPY_MODEL_4c3b6bc4f4974714b7f0916908c1dfc3","value":1979781432}},"3671af40a3944553bff108497c34d656":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_c114c204fec84c88be602aba53cefa27","placeholder":"","style":"IPY_MODEL_b7f6e0b0e9a5408ca8ab2a2c093c18bd","value":" 1.98G/1.98G [00:15<00:00, 179MB/s]"}},"ede441a2ff5a4970a9db369487dd91a6":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"92de8ec1a92a487eb872698185359a27":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"56dcdf86a9ad4342ac63a1d7c19bdbbd":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"fb49971dab3749bd854b68572fea4eb9":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"4c3b6bc4f4974714b7f0916908c1dfc3":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"c114c204fec84c88be602aba53cefa27":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"b7f6e0b0e9a5408ca8ab2a2c093c18bd":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"8950db7362044b0cb7a2df3f4a14ab96":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_51681304de6943ada122e5226a137a54","IPY_MODEL_f7ab29ccdaa14f20a9236f073d46cc5b","IPY_MODEL_d5d48ba443914e77b19dae2db44a2391"],"layout":"IPY_MODEL_0c163e318223420588abe6f5983417f2"}},"51681304de6943ada122e5226a137a54":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_33acdf4834b64baaa72dbb5c355d581f","placeholder":"","style":"IPY_MODEL_7c242694ca2e441bb23d67a1010051ba","value":"Downloading (…)of-00008.safetensors: 100%"}},"f7ab29ccdaa14f20a9236f073d46cc5b":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_f84bad5785e84fae84bdea480ecabe34","max":1946243984,"min":0,"orientation":"horizontal","style":"IPY_MODEL_6b6cc7afc2d54920950863c1eacba2d9","value":1946243984}},"d5d48ba443914e77b19dae2db44a2391":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_631b976952c44ee4ba5c3a580b554bc3","placeholder":"","style":"IPY_MODEL_40a2575e33c249d8983c96ab13ba1c87","value":" 1.95G/1.95G [00:14<00:00, 163MB/s]"}},"0c163e318223420588abe6f5983417f2":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"33acdf4834b64baaa72dbb5c355d581f":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"7c242694ca2e441bb23d67a1010051ba":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"f84bad5785e84fae84bdea480ecabe34":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"6b6cc7afc2d54920950863c1eacba2d9":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"631b976952c44ee4ba5c3a580b554bc3":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"40a2575e33c249d8983c96ab13ba1c87":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"64465e89b5084470a1300c230e4d6a0f":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_66f556e5045543d282cf6ec1b24e6a5d","IPY_MODEL_557fc212f965425d9cad706e31b5de71","IPY_MODEL_a3e402a256e1457caed5c73972707135"],"layout":"IPY_MODEL_ac237b12560c4d2a869ff7c4997034f6"}},"66f556e5045543d282cf6ec1b24e6a5d":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_9c477bace9474894a4af4a49ffb1b597","placeholder":"","style":"IPY_MODEL_1e94342bd4f34e16bba5277a1e287ea0","value":"Downloading (…)of-00008.safetensors: 100%"}},"557fc212f965425d9cad706e31b5de71":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_ee12e753e84b40f8b15a386726eeac74","max":1979781448,"min":0,"orientation":"horizontal","style":"IPY_MODEL_67759890afa7429bad93af68f98bca8c","value":1979781448}},"a3e402a256e1457caed5c73972707135":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_186d3c5f4a2a4ab3bb1ce46bdf4eaaa0","placeholder":"","style":"IPY_MODEL_e00ff0abba05463495cf89b4206de6a6","value":" 1.98G/1.98G [00:18<00:00, 159MB/s]"}},"ac237b12560c4d2a869ff7c4997034f6":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"9c477bace9474894a4af4a49ffb1b597":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"1e94342bd4f34e16bba5277a1e287ea0":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"ee12e753e84b40f8b15a386726eeac74":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"67759890afa7429bad93af68f98bca8c":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"186d3c5f4a2a4ab3bb1ce46bdf4eaaa0":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"e00ff0abba05463495cf89b4206de6a6":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"f662cc7314f04fd9b87b90dca6fe2948":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_f382b51901ba4384bc57bdb041c33fe8","IPY_MODEL_e2976b933e154328b6017b96e5a89f1b","IPY_MODEL_41a77c99fbb34fdd8e8d81cec7bd1e52"],"layout":"IPY_MODEL_60af0234884e4ba4bb3ca55357404002"}},"f382b51901ba4384bc57bdb041c33fe8":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_5db617b2eac744cab52546adb89f2b2d","placeholder":"","style":"IPY_MODEL_66bf1b86b6814d04bab8e602b4c6ab04","value":"Downloading (…)of-00008.safetensors: 100%"}},"e2976b933e154328b6017b96e5a89f1b":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_04881b0a754648a3b7b5d666c7140d3c","max":1946243984,"min":0,"orientation":"horizontal","style":"IPY_MODEL_2f6c0d3758834b5784101e9dbb76d8b6","value":1946243984}},"41a77c99fbb34fdd8e8d81cec7bd1e52":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_f703fe03badd4fd89769838d429633f6","placeholder":"","style":"IPY_MODEL_f5cffec9621847f3985a6d9309462c9f","value":" 1.95G/1.95G [00:15<00:00, 144MB/s]"}},"60af0234884e4ba4bb3ca55357404002":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"5db617b2eac744cab52546adb89f2b2d":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"66bf1b86b6814d04bab8e602b4c6ab04":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"04881b0a754648a3b7b5d666c7140d3c":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"2f6c0d3758834b5784101e9dbb76d8b6":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"f703fe03badd4fd89769838d429633f6":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"f5cffec9621847f3985a6d9309462c9f":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"3dd92a9f64d042f8a14d70a81b6b5041":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_6fc843a6157f424f8f48ded39a0b6174","IPY_MODEL_b9ad1baa021f471cbde376f215bb9553","IPY_MODEL_8f70605a96e54bbab6b1700917721305"],"layout":"IPY_MODEL_6be8aafb49574ea2b440d447cf068cc6"}},"6fc843a6157f424f8f48ded39a0b6174":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_d23ef9ab6fe34a67bae69675d6092b21","placeholder":"","style":"IPY_MODEL_2a581a92ccc24fb88091fe29ebb45dda","value":"Downloading (…)of-00008.safetensors: 100%"}},"b9ad1baa021f471cbde376f215bb9553":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_85073d535e8d4c698536ed2e4ccb49f0","max":1979781448,"min":0,"orientation":"horizontal","style":"IPY_MODEL_ac447978846b452abd3eadfa71d0acf5","value":1979781448}},"8f70605a96e54bbab6b1700917721305":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_30c37aa38bab4a6cb2cae9f7cc86c9bf","placeholder":"","style":"IPY_MODEL_2fd276b0f31449ca8cba6c8f641187e7","value":" 1.98G/1.98G [00:16<00:00, 153MB/s]"}},"6be8aafb49574ea2b440d447cf068cc6":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"d23ef9ab6fe34a67bae69675d6092b21":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"2a581a92ccc24fb88091fe29ebb45dda":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"85073d535e8d4c698536ed2e4ccb49f0":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"ac447978846b452abd3eadfa71d0acf5":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"30c37aa38bab4a6cb2cae9f7cc86c9bf":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"2fd276b0f31449ca8cba6c8f641187e7":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"aedc6763d2024505a03eaa3057d530f7":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_4fe5bede20774d0682c76c551e3fe4cf","IPY_MODEL_07d8f821c0654d068305616cc7e65169","IPY_MODEL_07e3343d7bb24d22ab54362f414c3f87"],"layout":"IPY_MODEL_e8c41663e554493e995d488b56acee6d"}},"4fe5bede20774d0682c76c551e3fe4cf":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_d2329f178afc4cac8314c5b120cd694e","placeholder":"","style":"IPY_MODEL_af362e7bd14345b2bab2d354de07a237","value":"Downloading (…)of-00008.safetensors: 100%"}},"07d8f821c0654d068305616cc7e65169":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_2d6a220aadf14400aefff2c3e63118f0","max":815834680,"min":0,"orientation":"horizontal","style":"IPY_MODEL_2fabde383f33481e87e46a3ca6462c6a","value":815834680}},"07e3343d7bb24d22ab54362f414c3f87":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_47c679a4a0774d9192ada75fb82ca5bd","placeholder":"","style":"IPY_MODEL_0f49cbe7bd0042a58c776be62e303f48","value":" 816M/816M [00:06<00:00, 136MB/s]"}},"e8c41663e554493e995d488b56acee6d":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"d2329f178afc4cac8314c5b120cd694e":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"af362e7bd14345b2bab2d354de07a237":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"2d6a220aadf14400aefff2c3e63118f0":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"2fabde383f33481e87e46a3ca6462c6a":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"47c679a4a0774d9192ada75fb82ca5bd":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"0f49cbe7bd0042a58c776be62e303f48":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"9f107fa9b45a403981e13e86ffadac33":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_ff835a6f8818434d9546e968621a40ac","IPY_MODEL_d95ca54538494e9c94d7baf831b8100f","IPY_MODEL_30f52c9064114bd29689c33374f35981"],"layout":"IPY_MODEL_59d4cb163ff5411d8fbf08d2ac7d6841"}},"ff835a6f8818434d9546e968621a40ac":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_ea027cd7d08643a49f2e7a9e6ad02c2c","placeholder":"","style":"IPY_MODEL_312daf1f2dee4b60a56d3c4c0373faec","value":"Loading checkpoint shards: 100%"}},"d95ca54538494e9c94d7baf831b8100f":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_a2a98e37c0054522b87236c3e4ad77e7","max":8,"min":0,"orientation":"horizontal","style":"IPY_MODEL_1bbd28564a09472d9c2e56ef538b5cfc","value":8}},"30f52c9064114bd29689c33374f35981":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_4a7d22479bee4e9da3fb14d3bdc1a9fe","placeholder":"","style":"IPY_MODEL_2aad5a8f58724b399bcadc2974270e87","value":" 8/8 [00:54<00:00, 4.88s/it]"}},"59d4cb163ff5411d8fbf08d2ac7d6841":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"ea027cd7d08643a49f2e7a9e6ad02c2c":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"312daf1f2dee4b60a56d3c4c0373faec":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"a2a98e37c0054522b87236c3e4ad77e7":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"1bbd28564a09472d9c2e56ef538b5cfc":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"4a7d22479bee4e9da3fb14d3bdc1a9fe":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"2aad5a8f58724b399bcadc2974270e87":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"6d0e1131adaa452e8e7e3552236c698a":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_9c69554d6cc4495aa5cc120ff5f0c1ef","IPY_MODEL_a06a11658ec34c779b92d1b545564bf8","IPY_MODEL_567b56683f2645539ade7409b0e2f8c2"],"layout":"IPY_MODEL_47fb621ad3e3441d8bda17e151f330bf"}},"9c69554d6cc4495aa5cc120ff5f0c1ef":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_4b5a4dc22f00478e88f65e97a90f7e16","placeholder":"","style":"IPY_MODEL_57be6290c12d4d7bb24de94802a69675","value":"Downloading (…)neration_config.json: 100%"}},"a06a11658ec34c779b92d1b545564bf8":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_b29459d2d8354e4c9f71a9f1e5a90b31","max":111,"min":0,"orientation":"horizontal","style":"IPY_MODEL_70cda0980f0f4044b2dfb60ae81c0be0","value":111}},"567b56683f2645539ade7409b0e2f8c2":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_78c04458e3ff4b9dac0c8bace9f3a6e8","placeholder":"","style":"IPY_MODEL_16d1fb7e26d7425680b960b5f48e6f24","value":" 111/111 [00:00<00:00, 1.84kB/s]"}},"47fb621ad3e3441d8bda17e151f330bf":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"4b5a4dc22f00478e88f65e97a90f7e16":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"57be6290c12d4d7bb24de94802a69675":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"b29459d2d8354e4c9f71a9f1e5a90b31":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"70cda0980f0f4044b2dfb60ae81c0be0":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"78c04458e3ff4b9dac0c8bace9f3a6e8":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"16d1fb7e26d7425680b960b5f48e6f24":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"c9d1aa1244f94da7a8bd51c6747979a6":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_04c9f4717ee34413be78241c08fca52d","IPY_MODEL_c7be18a542a341d488fe0e6787e0070a","IPY_MODEL_d40cd28139b24ab88a0f10fe8f397c61"],"layout":"IPY_MODEL_5444ab3413374fc8850e172d8991aeaf"}},"04c9f4717ee34413be78241c08fca52d":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_fe0319b1c543432b915c805cf8abaa73","placeholder":"","style":"IPY_MODEL_e53daa21f9ca4819970abb352a4b5fe5","value":"Downloading (…)okenizer_config.json: 100%"}},"c7be18a542a341d488fe0e6787e0070a":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_8700fbc216c2431583f30eac303f3113","max":1431,"min":0,"orientation":"horizontal","style":"IPY_MODEL_90cf63fb976040ee80d44172dcc6b313","value":1431}},"d40cd28139b24ab88a0f10fe8f397c61":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_ad75e13a28704a4982e7dc48ee7dfbb7","placeholder":"","style":"IPY_MODEL_6005fcdc7a354bb0bb86cddd8efc1b9e","value":" 1.43k/1.43k [00:00<00:00, 80.5kB/s]"}},"5444ab3413374fc8850e172d8991aeaf":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"fe0319b1c543432b915c805cf8abaa73":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"e53daa21f9ca4819970abb352a4b5fe5":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"8700fbc216c2431583f30eac303f3113":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"90cf63fb976040ee80d44172dcc6b313":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"ad75e13a28704a4982e7dc48ee7dfbb7":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"6005fcdc7a354bb0bb86cddd8efc1b9e":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"542495f1591d4d94b2d05e8d7d1f50fd":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_0012dac6fbec478a8e377a1652b937b8","IPY_MODEL_a2d02ed45bde4d8b8901b824eb488e2b","IPY_MODEL_eff08365f8e24b3083a7981ecc01c32a"],"layout":"IPY_MODEL_9c77d9fb38dd4b16b83e250159c294ee"}},"0012dac6fbec478a8e377a1652b937b8":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_786de9f41cda4cb38c080ae90b907794","placeholder":"","style":"IPY_MODEL_15f6b889c1d244b19239aed51bd7a06b","value":"Downloading tokenizer.model: 100%"}},"a2d02ed45bde4d8b8901b824eb488e2b":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_4a09188283a845ba980212eb22e27d7b","max":493443,"min":0,"orientation":"horizontal","style":"IPY_MODEL_179ffe54121a4f648da0ba61e25401d0","value":493443}},"eff08365f8e24b3083a7981ecc01c32a":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_9019d095a2e047ebbaad7ddddc742d39","placeholder":"","style":"IPY_MODEL_504447a3a3cf465cbbd6d35d2b02d8de","value":" 493k/493k [00:00<00:00, 30.7MB/s]"}},"9c77d9fb38dd4b16b83e250159c294ee":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"786de9f41cda4cb38c080ae90b907794":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"15f6b889c1d244b19239aed51bd7a06b":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"4a09188283a845ba980212eb22e27d7b":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"179ffe54121a4f648da0ba61e25401d0":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"9019d095a2e047ebbaad7ddddc742d39":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"504447a3a3cf465cbbd6d35d2b02d8de":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"169493fa776c4f32858d8ada28a9902e":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_012ba73ad3d7422e88f96dd7022ebcfe","IPY_MODEL_a8cebc6a039043b1a15672a65d2bc74c","IPY_MODEL_7e779d10cbb84fb6a3e154d593f5fd16"],"layout":"IPY_MODEL_c35f30e35de94ac2ae87fe1470613c68"}},"012ba73ad3d7422e88f96dd7022ebcfe":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_48a36dbdec3248fab8c9d915d9e53921","placeholder":"","style":"IPY_MODEL_83a8691b9c734409bf7caaf4ed21066a","value":"Downloading (…)/main/tokenizer.json: 100%"}},"a8cebc6a039043b1a15672a65d2bc74c":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_bfeb85b8bc944efb8afacad98676f8c2","max":1795303,"min":0,"orientation":"horizontal","style":"IPY_MODEL_72f9b6a11ea84512b326bced98e6e810","value":1795303}},"7e779d10cbb84fb6a3e154d593f5fd16":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_c26a277a40c04c099a830e0a346e8702","placeholder":"","style":"IPY_MODEL_845d26f25b9d4b0d945954d91c985a8a","value":" 1.80M/1.80M [00:00<00:00, 5.44MB/s]"}},"c35f30e35de94ac2ae87fe1470613c68":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"48a36dbdec3248fab8c9d915d9e53921":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"83a8691b9c734409bf7caaf4ed21066a":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"bfeb85b8bc944efb8afacad98676f8c2":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"72f9b6a11ea84512b326bced98e6e810":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"c26a277a40c04c099a830e0a346e8702":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"845d26f25b9d4b0d945954d91c985a8a":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"1c5cd44fd3104f6fa612e10ac432826d":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_caa327b8d84945eca6f4d733e617f3cc","IPY_MODEL_f6fa5f9808f1445aab7e726206749766","IPY_MODEL_7c4c81b536f44dc9914df6428c40ad1a"],"layout":"IPY_MODEL_e4aa69c959454612b19694f8f270d58c"}},"caa327b8d84945eca6f4d733e617f3cc":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_bc1a505eae3e42979cdbd30e87956445","placeholder":"","style":"IPY_MODEL_117c75b5f11348c1b8758fe4abacf503","value":"Downloading (…)in/added_tokens.json: 100%"}},"f6fa5f9808f1445aab7e726206749766":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_6502f05831104951b599c11867172ffc","max":42,"min":0,"orientation":"horizontal","style":"IPY_MODEL_6614c74c61b144bc91cd450e72ee93ff","value":42}},"7c4c81b536f44dc9914df6428c40ad1a":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_3037b05a942a464588ee7e54fa042334","placeholder":"","style":"IPY_MODEL_f65e3ce99b924989a8082a03b6cc2ef2","value":" 42.0/42.0 [00:00<00:00, 1.36kB/s]"}},"e4aa69c959454612b19694f8f270d58c":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"bc1a505eae3e42979cdbd30e87956445":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"117c75b5f11348c1b8758fe4abacf503":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"6502f05831104951b599c11867172ffc":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"6614c74c61b144bc91cd450e72ee93ff":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"3037b05a942a464588ee7e54fa042334":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"f65e3ce99b924989a8082a03b6cc2ef2":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"0d01fc2a5894448ab26771ce4f4ed0fc":{"model_module":"@jupyter-widgets/controls","model_name":"HBoxModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HBoxModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HBoxView","box_style":"","children":["IPY_MODEL_3792c555a4434615992afe39b5a168be","IPY_MODEL_7b25e155197f41ecba8d3018f56393a6","IPY_MODEL_ccb3b9ae2f0a4c4b957e2e03da01f1db"],"layout":"IPY_MODEL_397bb2e89c4e44e68467b24c342c7bf1"}},"3792c555a4434615992afe39b5a168be":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_1a15afd087b3413fbfc22b244c19d5ff","placeholder":"","style":"IPY_MODEL_f637d7f94ccc4ee18cac731bb337a92f","value":"Downloading (…)cial_tokens_map.json: 100%"}},"7b25e155197f41ecba8d3018f56393a6":{"model_module":"@jupyter-widgets/controls","model_name":"FloatProgressModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"FloatProgressModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"ProgressView","bar_style":"success","description":"","description_tooltip":null,"layout":"IPY_MODEL_881c7035aa0f4db5b690a7c9623c6e8c","max":168,"min":0,"orientation":"horizontal","style":"IPY_MODEL_5aff0a8982d247979886ae7951c9b224","value":168}},"ccb3b9ae2f0a4c4b957e2e03da01f1db":{"model_module":"@jupyter-widgets/controls","model_name":"HTMLModel","model_module_version":"1.5.0","state":{"_dom_classes":[],"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"HTMLModel","_view_count":null,"_view_module":"@jupyter-widgets/controls","_view_module_version":"1.5.0","_view_name":"HTMLView","description":"","description_tooltip":null,"layout":"IPY_MODEL_4b050fc85e0c405581ed55e850b183ab","placeholder":"","style":"IPY_MODEL_297c892dd4bf4886a62bcf4c6b4e3997","value":" 168/168 [00:00<00:00, 7.95kB/s]"}},"397bb2e89c4e44e68467b24c342c7bf1":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"1a15afd087b3413fbfc22b244c19d5ff":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"f637d7f94ccc4ee18cac731bb337a92f":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}},"881c7035aa0f4db5b690a7c9623c6e8c":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"5aff0a8982d247979886ae7951c9b224":{"model_module":"@jupyter-widgets/controls","model_name":"ProgressStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"ProgressStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","bar_color":null,"description_width":""}},"4b050fc85e0c405581ed55e850b183ab":{"model_module":"@jupyter-widgets/base","model_name":"LayoutModel","model_module_version":"1.2.0","state":{"_model_module":"@jupyter-widgets/base","_model_module_version":"1.2.0","_model_name":"LayoutModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"LayoutView","align_content":null,"align_items":null,"align_self":null,"border":null,"bottom":null,"display":null,"flex":null,"flex_flow":null,"grid_area":null,"grid_auto_columns":null,"grid_auto_flow":null,"grid_auto_rows":null,"grid_column":null,"grid_gap":null,"grid_row":null,"grid_template_areas":null,"grid_template_columns":null,"grid_template_rows":null,"height":null,"justify_content":null,"justify_items":null,"left":null,"margin":null,"max_height":null,"max_width":null,"min_height":null,"min_width":null,"object_fit":null,"object_position":null,"order":null,"overflow":null,"overflow_x":null,"overflow_y":null,"padding":null,"right":null,"top":null,"visibility":null,"width":null}},"297c892dd4bf4886a62bcf4c6b4e3997":{"model_module":"@jupyter-widgets/controls","model_name":"DescriptionStyleModel","model_module_version":"1.5.0","state":{"_model_module":"@jupyter-widgets/controls","_model_module_version":"1.5.0","_model_name":"DescriptionStyleModel","_view_count":null,"_view_module":"@jupyter-widgets/base","_view_module_version":"1.2.0","_view_name":"StyleView","description_width":""}}}}},"cells":[{"cell_type":"markdown","source":["# Quick start to run Zephyr 7B Alpha on Google Colab"],"metadata":{"id":"oSK6YKSx6H7m"}},{"cell_type":"code","source":["# Install transformers from source - only needed for versions <= v4.34\n","%pip install git+https://github.com/huggingface/transformers.git\n","%pip install accelerate"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"BLsaLpLP6QSK","executionInfo":{"status":"ok","timestamp":1697467249232,"user_tz":-120,"elapsed":45974,"user":{"displayName":"Lewis Tunstall","userId":"11991259263375974384"}},"outputId":"1b15ee88-5414-485f-9264-109e3d92d267"},"execution_count":1,"outputs":[{"output_type":"stream","name":"stdout","text":["Collecting git+https://github.com/huggingface/transformers.git\n"," Cloning https://github.com/huggingface/transformers.git to /tmp/pip-req-build-1as3weby\n"," Running command git clone --filter=blob:none --quiet https://github.com/huggingface/transformers.git /tmp/pip-req-build-1as3weby\n"," Resolved https://github.com/huggingface/transformers.git to commit 12cc1233591655528b3f8179c83a806de73fba3e\n"," Installing build dependencies ... \u001b[?25l\u001b[?25hdone\n"," Getting requirements to build wheel ... \u001b[?25l\u001b[?25hdone\n"," Preparing metadata (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n","Requirement already satisfied: filelock in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (3.12.4)\n","Collecting huggingface-hub<1.0,>=0.16.4 (from transformers==4.35.0.dev0)\n"," Downloading huggingface_hub-0.18.0-py3-none-any.whl (301 kB)\n","\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m302.0/302.0 kB\u001b[0m \u001b[31m5.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n","\u001b[?25hRequirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (1.23.5)\n","Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (23.2)\n","Requirement already satisfied: pyyaml>=5.1 in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (6.0.1)\n","Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (2023.6.3)\n","Requirement already satisfied: requests in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (2.31.0)\n","Collecting tokenizers<0.15,>=0.14 (from transformers==4.35.0.dev0)\n"," Downloading tokenizers-0.14.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (3.8 MB)\n","\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m3.8/3.8 MB\u001b[0m \u001b[31m19.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n","\u001b[?25hCollecting safetensors>=0.3.1 (from transformers==4.35.0.dev0)\n"," Downloading safetensors-0.4.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.3 MB)\n","\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.3/1.3 MB\u001b[0m \u001b[31m31.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n","\u001b[?25hRequirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.10/dist-packages (from transformers==4.35.0.dev0) (4.66.1)\n","Requirement already satisfied: fsspec>=2023.5.0 in /usr/local/lib/python3.10/dist-packages (from huggingface-hub<1.0,>=0.16.4->transformers==4.35.0.dev0) (2023.6.0)\n","Requirement already satisfied: typing-extensions>=3.7.4.3 in /usr/local/lib/python3.10/dist-packages (from huggingface-hub<1.0,>=0.16.4->transformers==4.35.0.dev0) (4.5.0)\n","Collecting huggingface-hub<1.0,>=0.16.4 (from transformers==4.35.0.dev0)\n"," Downloading huggingface_hub-0.17.3-py3-none-any.whl (295 kB)\n","\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m295.0/295.0 kB\u001b[0m \u001b[31m35.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n","\u001b[?25hRequirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python3.10/dist-packages (from requests->transformers==4.35.0.dev0) (3.3.0)\n","Requirement already satisfied: idna<4,>=2.5 in /usr/local/lib/python3.10/dist-packages (from requests->transformers==4.35.0.dev0) (3.4)\n","Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.10/dist-packages (from requests->transformers==4.35.0.dev0) (2.0.6)\n","Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.10/dist-packages (from requests->transformers==4.35.0.dev0) (2023.7.22)\n","Building wheels for collected packages: transformers\n"," Building wheel for transformers (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n"," Created wheel for transformers: filename=transformers-4.35.0.dev0-py3-none-any.whl size=7791033 sha256=443f1f957f04ac3efd24e0c6de6e5b07965a608ed56ae0da9b92b045af74a1c4\n"," Stored in directory: /tmp/pip-ephem-wheel-cache-5k_mcupq/wheels/e7/9c/5b/e1a9c8007c343041e61cc484433d512ea9274272e3fcbe7c16\n","Successfully built transformers\n","Installing collected packages: safetensors, huggingface-hub, tokenizers, transformers\n","Successfully installed huggingface-hub-0.17.3 safetensors-0.4.0 tokenizers-0.14.1 transformers-4.35.0.dev0\n","Collecting accelerate\n"," Downloading accelerate-0.23.0-py3-none-any.whl (258 kB)\n","\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m258.1/258.1 kB\u001b[0m \u001b[31m4.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n","\u001b[?25hRequirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.10/dist-packages (from accelerate) (1.23.5)\n","Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.10/dist-packages (from accelerate) (23.2)\n","Requirement already satisfied: psutil in /usr/local/lib/python3.10/dist-packages (from accelerate) (5.9.5)\n","Requirement already satisfied: pyyaml in /usr/local/lib/python3.10/dist-packages (from accelerate) (6.0.1)\n","Requirement already satisfied: torch>=1.10.0 in /usr/local/lib/python3.10/dist-packages (from accelerate) (2.0.1+cu118)\n","Requirement already satisfied: huggingface-hub in /usr/local/lib/python3.10/dist-packages (from accelerate) (0.17.3)\n","Requirement already satisfied: filelock in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (3.12.4)\n","Requirement already satisfied: typing-extensions in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (4.5.0)\n","Requirement already satisfied: sympy in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (1.12)\n","Requirement already satisfied: networkx in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (3.1)\n","Requirement already satisfied: jinja2 in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (3.1.2)\n","Requirement already satisfied: triton==2.0.0 in /usr/local/lib/python3.10/dist-packages (from torch>=1.10.0->accelerate) (2.0.0)\n","Requirement already satisfied: cmake in /usr/local/lib/python3.10/dist-packages (from triton==2.0.0->torch>=1.10.0->accelerate) (3.27.6)\n","Requirement already satisfied: lit in /usr/local/lib/python3.10/dist-packages (from triton==2.0.0->torch>=1.10.0->accelerate) (17.0.2)\n","Requirement already satisfied: fsspec in /usr/local/lib/python3.10/dist-packages (from huggingface-hub->accelerate) (2023.6.0)\n","Requirement already satisfied: requests in /usr/local/lib/python3.10/dist-packages (from huggingface-hub->accelerate) (2.31.0)\n","Requirement already satisfied: tqdm>=4.42.1 in /usr/local/lib/python3.10/dist-packages (from huggingface-hub->accelerate) (4.66.1)\n","Requirement already satisfied: MarkupSafe>=2.0 in /usr/local/lib/python3.10/dist-packages (from jinja2->torch>=1.10.0->accelerate) (2.1.3)\n","Requirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python3.10/dist-packages (from requests->huggingface-hub->accelerate) (3.3.0)\n","Requirement already satisfied: idna<4,>=2.5 in /usr/local/lib/python3.10/dist-packages (from requests->huggingface-hub->accelerate) (3.4)\n","Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.10/dist-packages (from requests->huggingface-hub->accelerate) (2.0.6)\n","Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.10/dist-packages (from requests->huggingface-hub->accelerate) (2023.7.22)\n","Requirement already satisfied: mpmath>=0.19 in /usr/local/lib/python3.10/dist-packages (from sympy->torch>=1.10.0->accelerate) (1.3.0)\n","Installing collected packages: accelerate\n","Successfully installed accelerate-0.23.0\n"]}]},{"cell_type":"markdown","source":["## Load dependencies"],"metadata":{"id":"YK6Ux0Xy7D4o"}},{"cell_type":"code","source":["import torch\n","from transformers import pipeline"],"metadata":{"id":"fLjPZ5ke6q8Y","executionInfo":{"status":"ok","timestamp":1697467265528,"user_tz":-120,"elapsed":15540,"user":{"displayName":"Lewis Tunstall","userId":"11991259263375974384"}}},"execution_count":2,"outputs":[]},{"cell_type":"markdown","source":["## Download model and load pipeline"],"metadata":{"id":"e1mBWABw7MAQ"}},{"cell_type":"code","source":["pipe = pipeline(\"text-generation\", model=\"HuggingFaceH4/zephyr-7b-alpha\", torch_dtype=torch.bfloat16, device_map=\"auto\")"],"metadata":{"colab":{"base_uri":"https://localhost:8080/","height":610,"referenced_widgets":["af35743bd7cc4b2a9a9ac89166372f1f","cc8496246b9248ae8631168804de7374","0a9a7266561e40a991473c40721d1caa","83bb42e8a79446cfa6bdfb76284a2f2c","a3df2076447a4f5e8b777295f55c8fb4","cb54c0578558402bb0b8341999449b92","a76ea86b1887442e8828d1184f548b05","9659515b9831421d9464ffee29b2b87a","a08f5e40fb1e4662aa624e6feb3eddac","790926668e1e4c4ab390c0821529c3c1","d3b27018eaf34defa08ff8c0c09ef826","2f7c528ac4a846b3bd3f5b859c5937ce","779303ae3a504bc39e94c05bb41d6680","12b3162cbf9c4c3ba4f83f325341fc26","ed0a166c01104946b3768f827f9a92da","99f491f558e248abb4c7acc9deb28923","91eaa2a5525f4f51a111ad59dca862dd","70c24596219147189534ef2effd75212","e54d337cbae84671ba46d2fccf7241a5","586b2f72d4694e3c88f22131326378b6","3abcf30b8b7a49d480d80b43f7528f3b","030709c0956f400c9c61d9c5028a7b5a","c2e47ceae1d14da59867cc5653409e7e","07b3f2f9ab8e479489cfa96d8ced3f4c","5bd6352bb7bf4c2fa195c73d9b875ea7","02eab2611e47411eb957e42321093df2","4dec5009208844fc845d276f86c53ea2","cdde97c62f854f8f96a86957c4e67487","1fa0cf1349ed4f779a3f61f8f5be0f1e","f28f325a3aa24182856e5da82b565ddd","56e4af5782d54e9c8069444b34fc6b79","bbad101d94d64d3e8f47f2d22bf3096e","33f0fd3841ac42788e92041f2756df4a","6cf085e28dc444698da1fcb0671d5570","1d2644bfdfa1493bb682f9bf80e132be","d28ec950aa0d4c5b817e54172da5fb44","db47d15652ac4a51b85cbb0e1f49cffc","6148302fcdcd4eeea8770340bcdc4af5","09c91af5582f40d1b864a3f1a6728190","a3cc8c7cdec74aa3a42ae888fd12b813","0977d1f5e8f54ee1ae13509423b6a530","5695943f70204d65aac3d6a994fcac9c","8a2f7e2ba8254dd8929538d88c1be72e","e5981daa2da24c299dec59442d3cd3d9","5660194f98bd47f9a872fb3ba194e722","287f1b8059ff461580964d6765505048","6106c56288dc46fe812176811ede8154","faf9c3ef160d4772b6b4d74b928f8df8","ba143a50545a4c5588ffa381b3d86457","9d06d6845ae84d3ebfc61d1414c141ab","baae3ea716dc4c26bc1720f25b2eee60","c667c1db826741f0ae7e8954071b953e","89f63cd121554cbbb201cb999ac17515","6e0d6d6061924a9080e46924e0a185d5","4a73e20d2402485fa864b16e0cfe40e8","ebf0e040eeee4b2896f03f4c6eb2b297","b235c4229aa943759638c05f7ee365ce","111c3f38704d432b828946a63c909381","3671af40a3944553bff108497c34d656","ede441a2ff5a4970a9db369487dd91a6","92de8ec1a92a487eb872698185359a27","56dcdf86a9ad4342ac63a1d7c19bdbbd","fb49971dab3749bd854b68572fea4eb9","4c3b6bc4f4974714b7f0916908c1dfc3","c114c204fec84c88be602aba53cefa27","b7f6e0b0e9a5408ca8ab2a2c093c18bd","8950db7362044b0cb7a2df3f4a14ab96","51681304de6943ada122e5226a137a54","f7ab29ccdaa14f20a9236f073d46cc5b","d5d48ba443914e77b19dae2db44a2391","0c163e318223420588abe6f5983417f2","33acdf4834b64baaa72dbb5c355d581f","7c242694ca2e441bb23d67a1010051ba","f84bad5785e84fae84bdea480ecabe34","6b6cc7afc2d54920950863c1eacba2d9","631b976952c44ee4ba5c3a580b554bc3","40a2575e33c249d8983c96ab13ba1c87","64465e89b5084470a1300c230e4d6a0f","66f556e5045543d282cf6ec1b24e6a5d","557fc212f965425d9cad706e31b5de71","a3e402a256e1457caed5c73972707135","ac237b12560c4d2a869ff7c4997034f6","9c477bace9474894a4af4a49ffb1b597","1e94342bd4f34e16bba5277a1e287ea0","ee12e753e84b40f8b15a386726eeac74","67759890afa7429bad93af68f98bca8c","186d3c5f4a2a4ab3bb1ce46bdf4eaaa0","e00ff0abba05463495cf89b4206de6a6","f662cc7314f04fd9b87b90dca6fe2948","f382b51901ba4384bc57bdb041c33fe8","e2976b933e154328b6017b96e5a89f1b","41a77c99fbb34fdd8e8d81cec7bd1e52","60af0234884e4ba4bb3ca55357404002","5db617b2eac744cab52546adb89f2b2d","66bf1b86b6814d04bab8e602b4c6ab04","04881b0a754648a3b7b5d666c7140d3c","2f6c0d3758834b5784101e9dbb76d8b6","f703fe03badd4fd89769838d429633f6","f5cffec9621847f3985a6d9309462c9f","3dd92a9f64d042f8a14d70a81b6b5041","6fc843a6157f424f8f48ded39a0b6174","b9ad1baa021f471cbde376f215bb9553","8f70605a96e54bbab6b1700917721305","6be8aafb49574ea2b440d447cf068cc6","d23ef9ab6fe34a67bae69675d6092b21","2a581a92ccc24fb88091fe29ebb45dda","85073d535e8d4c698536ed2e4ccb49f0","ac447978846b452abd3eadfa71d0acf5","30c37aa38bab4a6cb2cae9f7cc86c9bf","2fd276b0f31449ca8cba6c8f641187e7","aedc6763d2024505a03eaa3057d530f7","4fe5bede20774d0682c76c551e3fe4cf","07d8f821c0654d068305616cc7e65169","07e3343d7bb24d22ab54362f414c3f87","e8c41663e554493e995d488b56acee6d","d2329f178afc4cac8314c5b120cd694e","af362e7bd14345b2bab2d354de07a237","2d6a220aadf14400aefff2c3e63118f0","2fabde383f33481e87e46a3ca6462c6a","47c679a4a0774d9192ada75fb82ca5bd","0f49cbe7bd0042a58c776be62e303f48","9f107fa9b45a403981e13e86ffadac33","ff835a6f8818434d9546e968621a40ac","d95ca54538494e9c94d7baf831b8100f","30f52c9064114bd29689c33374f35981","59d4cb163ff5411d8fbf08d2ac7d6841","ea027cd7d08643a49f2e7a9e6ad02c2c","312daf1f2dee4b60a56d3c4c0373faec","a2a98e37c0054522b87236c3e4ad77e7","1bbd28564a09472d9c2e56ef538b5cfc","4a7d22479bee4e9da3fb14d3bdc1a9fe","2aad5a8f58724b399bcadc2974270e87","6d0e1131adaa452e8e7e3552236c698a","9c69554d6cc4495aa5cc120ff5f0c1ef","a06a11658ec34c779b92d1b545564bf8","567b56683f2645539ade7409b0e2f8c2","47fb621ad3e3441d8bda17e151f330bf","4b5a4dc22f00478e88f65e97a90f7e16","57be6290c12d4d7bb24de94802a69675","b29459d2d8354e4c9f71a9f1e5a90b31","70cda0980f0f4044b2dfb60ae81c0be0","78c04458e3ff4b9dac0c8bace9f3a6e8","16d1fb7e26d7425680b960b5f48e6f24","c9d1aa1244f94da7a8bd51c6747979a6","04c9f4717ee34413be78241c08fca52d","c7be18a542a341d488fe0e6787e0070a","d40cd28139b24ab88a0f10fe8f397c61","5444ab3413374fc8850e172d8991aeaf","fe0319b1c543432b915c805cf8abaa73","e53daa21f9ca4819970abb352a4b5fe5","8700fbc216c2431583f30eac303f3113","90cf63fb976040ee80d44172dcc6b313","ad75e13a28704a4982e7dc48ee7dfbb7","6005fcdc7a354bb0bb86cddd8efc1b9e","542495f1591d4d94b2d05e8d7d1f50fd","0012dac6fbec478a8e377a1652b937b8","a2d02ed45bde4d8b8901b824eb488e2b","eff08365f8e24b3083a7981ecc01c32a","9c77d9fb38dd4b16b83e250159c294ee","786de9f41cda4cb38c080ae90b907794","15f6b889c1d244b19239aed51bd7a06b","4a09188283a845ba980212eb22e27d7b","179ffe54121a4f648da0ba61e25401d0","9019d095a2e047ebbaad7ddddc742d39","504447a3a3cf465cbbd6d35d2b02d8de","169493fa776c4f32858d8ada28a9902e","012ba73ad3d7422e88f96dd7022ebcfe","a8cebc6a039043b1a15672a65d2bc74c","7e779d10cbb84fb6a3e154d593f5fd16","c35f30e35de94ac2ae87fe1470613c68","48a36dbdec3248fab8c9d915d9e53921","83a8691b9c734409bf7caaf4ed21066a","bfeb85b8bc944efb8afacad98676f8c2","72f9b6a11ea84512b326bced98e6e810","c26a277a40c04c099a830e0a346e8702","845d26f25b9d4b0d945954d91c985a8a","1c5cd44fd3104f6fa612e10ac432826d","caa327b8d84945eca6f4d733e617f3cc","f6fa5f9808f1445aab7e726206749766","7c4c81b536f44dc9914df6428c40ad1a","e4aa69c959454612b19694f8f270d58c","bc1a505eae3e42979cdbd30e87956445","117c75b5f11348c1b8758fe4abacf503","6502f05831104951b599c11867172ffc","6614c74c61b144bc91cd450e72ee93ff","3037b05a942a464588ee7e54fa042334","f65e3ce99b924989a8082a03b6cc2ef2","0d01fc2a5894448ab26771ce4f4ed0fc","3792c555a4434615992afe39b5a168be","7b25e155197f41ecba8d3018f56393a6","ccb3b9ae2f0a4c4b957e2e03da01f1db","397bb2e89c4e44e68467b24c342c7bf1","1a15afd087b3413fbfc22b244c19d5ff","f637d7f94ccc4ee18cac731bb337a92f","881c7035aa0f4db5b690a7c9623c6e8c","5aff0a8982d247979886ae7951c9b224","4b050fc85e0c405581ed55e850b183ab","297c892dd4bf4886a62bcf4c6b4e3997"]},"id":"XMeitcMd7G-a","executionInfo":{"status":"ok","timestamp":1697467460146,"user_tz":-120,"elapsed":194632,"user":{"displayName":"Lewis Tunstall","userId":"11991259263375974384"}},"outputId":"c71db626-b8b0-471a-84a1-6e3c6e0e6179"},"execution_count":3,"outputs":[{"output_type":"display_data","data":{"text/plain":["Downloading (…)lve/main/config.json: 0%| | 0.00/628 [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"af35743bd7cc4b2a9a9ac89166372f1f"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)fetensors.index.json: 0%| | 0.00/23.9k [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"2f7c528ac4a846b3bd3f5b859c5937ce"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading shards: 0%| | 0/8 [00:00, ?it/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"c2e47ceae1d14da59867cc5653409e7e"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.89G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"6cf085e28dc444698da1fcb0671d5570"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.95G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"5660194f98bd47f9a872fb3ba194e722"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.98G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"ebf0e040eeee4b2896f03f4c6eb2b297"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.95G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"8950db7362044b0cb7a2df3f4a14ab96"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.98G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"64465e89b5084470a1300c230e4d6a0f"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.95G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"f662cc7314f04fd9b87b90dca6fe2948"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/1.98G [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"3dd92a9f64d042f8a14d70a81b6b5041"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)of-00008.safetensors: 0%| | 0.00/816M [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"aedc6763d2024505a03eaa3057d530f7"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Loading checkpoint shards: 0%| | 0/8 [00:00, ?it/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"9f107fa9b45a403981e13e86ffadac33"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)neration_config.json: 0%| | 0.00/111 [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"6d0e1131adaa452e8e7e3552236c698a"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)okenizer_config.json: 0%| | 0.00/1.43k [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"c9d1aa1244f94da7a8bd51c6747979a6"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading tokenizer.model: 0%| | 0.00/493k [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"542495f1591d4d94b2d05e8d7d1f50fd"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)/main/tokenizer.json: 0%| | 0.00/1.80M [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"169493fa776c4f32858d8ada28a9902e"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)in/added_tokens.json: 0%| | 0.00/42.0 [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"1c5cd44fd3104f6fa612e10ac432826d"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":["Downloading (…)cial_tokens_map.json: 0%| | 0.00/168 [00:00, ?B/s]"],"application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"0d01fc2a5894448ab26771ce4f4ed0fc"}},"metadata":{}},{"output_type":"stream","name":"stderr","text":["Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"]}]},{"cell_type":"markdown","source":["## Prepare inputs"],"metadata":{"id":"heIH5UYR71Ui"}},{"cell_type":"code","source":["# Each message can have 1 of 3 roles: \"system\" (to provide initial instructions), \"user\", or \"assistant\". For inference, make sure \"user\" is the role in the final message.\n","messages = [\n"," {\n"," \"role\": \"system\",\n"," \"content\": \"You are a friendly chatbot who always responds in the style of a pirate.\",\n"," },\n"," {\"role\": \"user\", \"content\": \"How many helicopters can a human eat in one sitting?\"},\n","]\n","# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating\n","prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)\n","print(prompt)"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"UKcD_fy_7QYD","executionInfo":{"status":"ok","timestamp":1697467470454,"user_tz":-120,"elapsed":7,"user":{"displayName":"Lewis Tunstall","userId":"11991259263375974384"}},"outputId":"2d06cb56-dd95-42b9-e963-eece727a53cd"},"execution_count":4,"outputs":[{"output_type":"stream","name":"stderr","text":["Using sep_token, but it is not set yet.\n","Using cls_token, but it is not set yet.\n","Using mask_token, but it is not set yet.\n"]},{"output_type":"stream","name":"stdout","text":["<|system|>\n","You are a friendly chatbot who always responds in the style of a pirate\n","<|user|>\n","How many helicopters can a human eat in one sitting?\n","<|assistant|>\n","\n"]}]},{"cell_type":"markdown","source":["## Generate!"],"metadata":{"id":"20K8tYJo7_7b"}},{"cell_type":"code","source":["outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)\n","print(outputs[0][\"generated_text\"])"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"xc8RZR0b8Br2","executionInfo":{"status":"ok","timestamp":1697467523847,"user_tz":-120,"elapsed":47326,"user":{"displayName":"Lewis Tunstall","userId":"11991259263375974384"}},"outputId":"55bb1c98-345f-4d6b-9a31-a5b1560ab3d2"},"execution_count":5,"outputs":[{"output_type":"stream","name":"stderr","text":["/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py:1462: UserWarning: You have modified the pretrained model configuration to control generation. This is a deprecated strategy to control generation and will be removed soon, in a future version. Please use and modify the model generation configuration (see https://huggingface.co/docs/transformers/generation_strategies#default-text-generation-configuration )\n"," warnings.warn(\n"]},{"output_type":"stream","name":"stdout","text":["<|system|>\n","You are a friendly chatbot who always responds in the style of a pirate\n","<|user|>\n","How many helicopters can a human eat in one sitting?\n","<|assistant|>\n","Ahoy matey! Me be thinkin' that a human can't eat a helicopter, for it's not food, it's a flying machine. Me be confused by your question, but me can assure ye that a human can't eat a helicopter in one sitting!\n"]}]},{"cell_type":"code","source":[],"metadata":{"id":"OJKvdt1B8if0"},"execution_count":null,"outputs":[]}]}
\ No newline at end of file
diff --git a/config.json b/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..0d8f4f684f74444a9371d2359830368851268357
--- /dev/null
+++ b/config.json
@@ -0,0 +1,26 @@
+{
+ "_name_or_path": "./zephyr-7b-alpha/",
+ "architectures": [
+ "MistralForCausalLM"
+ ],
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "hidden_act": "silu",
+ "hidden_size": 4096,
+ "initializer_range": 0.02,
+ "intermediate_size": 14336,
+ "max_position_embeddings": 32768,
+ "model_type": "mistral",
+ "num_attention_heads": 32,
+ "num_hidden_layers": 32,
+ "num_key_value_heads": 8,
+ "pad_token_id": 2,
+ "rms_norm_eps": 1e-05,
+ "rope_theta": 10000.0,
+ "sliding_window": 4096,
+ "tie_word_embeddings": false,
+ "torch_dtype": "bfloat16",
+ "transformers_version": "4.34.0",
+ "use_cache": true,
+ "vocab_size": 32000
+}
diff --git a/eval_results.json b/eval_results.json
new file mode 100644
index 0000000000000000000000000000000000000000..ebfacdfda5fb96c0b250fefbffc79e60bd1a50f5
--- /dev/null
+++ b/eval_results.json
@@ -0,0 +1,16 @@
+{
+ "epoch": 1.0,
+ "eval_logits/chosen": -2.744652509689331,
+ "eval_logits/rejected": -2.71529483795166,
+ "eval_logps/chosen": -297.10400390625,
+ "eval_logps/rejected": -327.4286193847656,
+ "eval_loss": 0.46045970916748047,
+ "eval_rewards/accuracies": 0.78125,
+ "eval_rewards/chosen": -0.5052940249443054,
+ "eval_rewards/margins": 1.3699172735214233,
+ "eval_rewards/rejected": -1.8752113580703735,
+ "eval_runtime": 52.3218,
+ "eval_samples": 1000,
+ "eval_samples_per_second": 19.113,
+ "eval_steps_per_second": 0.306
+}
\ No newline at end of file
diff --git a/generation_config.json b/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..7327bd306f5194519214f3480ea06c6dcd8ceefe
--- /dev/null
+++ b/generation_config.json
@@ -0,0 +1,6 @@
+{
+ "_from_model_config": true,
+ "bos_token_id": 1,
+ "eos_token_id": 2,
+ "transformers_version": "4.34.0"
+}
diff --git a/input_states.safetensors b/input_states.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..30166efedf568b28daebf34304e3b77ba7a29734
--- /dev/null
+++ b/input_states.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ee900988e963f08ba20362e94528b29970774c3bea7240ebfd7e294682c9777
+size 1677721696
diff --git a/job.json b/job.json
new file mode 100644
index 0000000000000000000000000000000000000000..057b39690b9bc187f71abf5785ba9d72c39a44ba
--- /dev/null
+++ b/job.json
@@ -0,0 +1,103424 @@
+{
+ "in_dir": "/content/zephyr-7b-alpha",
+ "out_dir": "/content/zephyr_alpha_quantized",
+ "cal_dataset": "/content/wikitext-test.parquet",
+ "dataset_rows": 100,
+ "measurement_rows": 16,
+ "gpu_rows": 0,
+ "length": 2048,
+ "measurement_length": 2048,
+ "bits": 4.0,
+ "head_bits": 6,
+ "progress": "quant",
+ "shard_size": 8192,
+ "output_measurement": null,
+ "compile_full": null,
+ "rope_scale": 1.0,
+ "rope_alpha": 1.0,
+ "cal_filename": "/content/zephyr_alpha_quantized/cal_data.safetensors",
+ "last_module_idx": 66,
+ "measurement": [
+ {
+ "key": "model.layers.0.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.011168720200657845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.009599272161722183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.004790821112692356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.005077867768704891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0050776307471096516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0020578650292009115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.01079169474542141,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.009518211707472801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.005259049125015736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.004644699394702911,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0048562814481556416,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0050615472719073296,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.004643067717552185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0027538840658962727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.002120051998645067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0026784869842231274,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.001900350907817483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.001615153276361525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0018467125482857227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.0015654281014576554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0017823567613959312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0018463897285982966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0014250442618504167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.001544640981592238,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.011168720200657845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.011168720200657845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.011349556967616081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.00980227068066597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.004773853346705437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.005033744964748621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.005033500958234072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.001928701065480709,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.01110624335706234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.009691532701253891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.005209838971495628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.004562461748719215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0047759003937244415,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.005032561253756285,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.004560141358524561,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.002655236516147852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.00194216996897012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0026120387483388186,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0016779541037976742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.001346124685369432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0016128767747431993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0012812796048820019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.001617392641492188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0016132771270349622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0012200467754155397,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0012539472663775086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.011349556967616081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.011349556967616081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.11578802764415741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07042404264211655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.043504536151885986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.050045717507600784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0499984472990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.024094626307487488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.07798509299755096,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.06458660215139389,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0545671209692955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03145388513803482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03755158931016922,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.042070720344781876,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.031096184626221657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02426334097981453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022416630759835243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02171494998037815,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01284871157258749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011072240769863129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009356767870485783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007840869016945362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.011130761355161667,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.009300582110881805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007214382756501436,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.006086327601224184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.043504536151885986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.043504536151885986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11313514411449432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0732056200504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05039582401514053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04979194700717926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.048297327011823654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.027966083958745003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07069999724626541,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06314126402139664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.053515732288360596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.031263019889593124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03367261961102486,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03614491969347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030515408143401146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024038290604948997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022272668778896332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0183316171169281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013834427110850811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012738020159304142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011082837358117104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009938649833202362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01025769766420126,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.010982340201735497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00825857650488615,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0086810402572155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05039582401514053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05039582401514053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.11327356100082397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1058005541563034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.10340237617492676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.09472587704658508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.051241014152765274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.04909159988164902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.05705980956554413,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.052599381655454636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.051703788340091705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.046595051884651184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.04498507082462311,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.029025588184595108,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.02526126801967621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.024675775319337845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02453121356666088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.01461299229413271,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.013298182748258114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.013174419291317463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.012456698343157768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.012386062182486057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.008203603327274323,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.008870589546859264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.008029122836887836,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.006887989118695259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.052599381655454636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.052599381655454636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.13404832780361176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.12670819461345673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12447718530893326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.114055335521698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06098993122577667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.05894749239087105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.06735212355852127,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.06225910410284996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06142910569906235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.055878788232803345,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05380960926413536,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03398489952087402,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.029508553445339203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.02898118458688259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02886049821972847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.016938773915171623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01486075110733509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.014734474010765553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01386601198464632,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.013790595345199108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.008871518075466156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.008892863988876343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.008702414110302925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.005870731081813574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05380960926413536,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05380960926413536,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.0.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.07670606672763824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.06584762781858444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.06038632243871689,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.05355939269065857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.03438195586204529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.029740920290350914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.04498032480478287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.03952185437083244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.03574554622173309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.028567753732204437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.02737824246287346,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.022440247237682343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.019085902720689774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.01691751927137375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.016391701996326447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.011539027094841003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.00958372000604868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.009348219260573387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.008578285574913025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.008279938250780106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.006880487315356731,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.00712636299431324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.00626949081197381,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0057810521684587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.05355939269065857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.05355939269065857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.019663630053400993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.013406879268586636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.007660592906177044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.008349855430424213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.00809794757515192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0036465248558670282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.013933522626757622,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.012625516392290592,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0092203663662076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.006169123109430075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.006603498011827469,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.007054396439343691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.006034728605300188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0040495740249753,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.003436762373894453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0035601980052888393,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0023379670456051826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.0019536123145371675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0020181608851999044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.001642401795834303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0019508769037202,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0019979930948466063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0013339562574401498,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0014734480064362288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.019663630053400993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.019663630053400993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.017518868669867516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.01219908520579338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.006678944453597069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.007290045265108347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0070999860763549805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0029782995115965605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.013018052093684673,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.011648166924715042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.008138769306242466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.005624793935567141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.006064086686819792,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0065219649113714695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.005529758520424366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0035434754099696875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.0028902385383844376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.003280432429164648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0020198565907776356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0016353140817955136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0017707019578665495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0013762976741418242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0017502045957371593,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0017561280401423573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.001087298383936286,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0012416673125699162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.017518868669867516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.017518868669867516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.13320785760879517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08857599645853043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06414425373077393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.062488194555044174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.057051729410886765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.034113503992557526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.08158433437347412,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.07376103848218918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0631725862622261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03857894614338875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.039744194597005844,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04156964644789696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.03522882238030434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.027669740840792656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.02555888704955578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.020771265029907227,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.014558087103068829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.012766730040311813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011037899181246758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009418193250894547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.010778137482702732,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.010443516075611115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.008073403500020504,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.006722630932927132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03857894614338875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03857894614338875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.15270331501960754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.12318339943885803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.11086562275886536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09390351921319962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07004237920045853,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05806812644004822,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.08862987905740738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.07844515144824982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07241526991128922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05164502561092377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.049597982317209244,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04550931602716446,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0382709726691246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.034586384892463684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.033655546605587006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02309839054942131,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01936962828040123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01856229268014431,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01617896556854248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.015599473379552364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013060690835118294,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014155326411128044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.011925097554922104,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011372439563274384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05164502561092377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05164502561092377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1567811667919159,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1485101878643036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14568355679512024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13279618322849274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0726742148399353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07009439915418625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0808241069316864,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0739935040473938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07315933704376221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06607392430305481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06361325085163116,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04134838655591011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03595521301031113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03534533828496933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03519531339406967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021018361672759056,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019455352798104286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019302669912576675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01826331578195095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01818668842315674,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012167708948254585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013381781987845898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012010962702333927,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010793034918606281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04134838655591011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04134838655591011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.19255827367305756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.18272824585437775,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.17977085709571838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.16407370567321777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08971576392650604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.08672579377889633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.098969966173172,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.09122589230537415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09027554839849472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08172681927680969,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0785735547542572,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.050545584410429,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.043932583183050156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04328686743974686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04312015324831009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.025338059291243553,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02311822585761547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02295522764325142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02161826193332672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02152479812502861,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.013921397738158703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015096379444003105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.013726180419325829,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011499397456645966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.050545584410429,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.050545584410429,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.1.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.019199078902602196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.01878763735294342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.006921666674315929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.00646975776180625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.005650496110320091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0034061234910041094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.018811136484146118,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.018254732713103294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.005731276702135801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.005354026332497597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0053440118208527565,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.005248498171567917,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.005061940755695105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0026619865093380213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.0026766916271299124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.002273819176480174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.0023983637802302837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.001029878854751587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.0023623292800039053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.000951776746660471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0023449964355677366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.0023288021329790354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.0007354797562584281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0007976787746883929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.019199078902602196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.019199078902602196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.05501016974449158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.043334461748600006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.03665383905172348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.03312649205327034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.024345438927412033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.018490733578801155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.03337150067090988,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03038576804101467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.025987472385168076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.018868857994675636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.018617024645209312,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.01694329082965851,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.014517929404973984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.011795316822826862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.011069130152463913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.008465803228318691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.00624634325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.005706323776394129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.005192603450268507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.004691623616963625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00441138306632638,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.004422945436090231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0034645930863916874,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0029624789021909237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.043334461748600006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.043334461748600006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.05672017112374306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04279809445142746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.03405015915632248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.03148718550801277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.024548310786485672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.017191600054502487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03530892729759216,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.032226257026195526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.026598580181598663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.01862349361181259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.018713288009166718,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.017903584986925125,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.015325364656746387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.011925559490919113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.010991599410772324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.008935020305216312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0063719660975039005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.005681759677827358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0052579911425709724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.004604979418218136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.004677099175751209,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004683797247707844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0035144537687301636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00315110944211483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04279809445142746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04279809445142746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.16881753504276276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1410287767648697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.12958964705467224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.11464769393205643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.07729382812976837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.06546717882156372,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09526020288467407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08619674295186996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08046350628137589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.061131030321121216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.058465443551540375,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04843978211283684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.041259944438934326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.03719085827469826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03618834540247917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02420763485133648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01911242865025997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.018178023397922516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.015842380002141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.015133141539990902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012576712295413017,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012094818986952305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011064727790653706,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0076539963483810425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04843978211283684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04843978211283684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.156541645526886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.13858985900878906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.12967531383037567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.11246905475854874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07320044189691544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.06499434262514114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09032202512025833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.08170590549707413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07497423142194748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06085894629359245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05692308768630028,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04705378785729408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0400383360683918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.036137260496616364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.035172898322343826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024017736315727234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.020189229398965836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.019475512206554413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0179683156311512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.017397213727235794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013736758381128311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014681994915008545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01252848468720913,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01172161940485239,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04705378785729408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04705378785729408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.19975252449512482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.18823857605457306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1846926212310791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.16731983423233032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09383014589548111,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09011030942201614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.10378500074148178,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.09576664119958878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09457574784755707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08421718329191208,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08041726052761078,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05310473218560219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04596571624279022,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04511168226599693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04491891711950302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02657609060406685,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.023638196289539337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.023406749591231346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.021765321493148804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.021637681871652603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.014316429384052753,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.014850785955786705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.014048966579139233,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010670391842722893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05310473218560219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05310473218560219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23413872718811035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22090375423431396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21686704456806183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19659093022346497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10993479937314987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10565081238746643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12136806547641754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.1120128333568573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11077161878347397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09870945662260056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0940827876329422,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06181109696626663,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05351909250020981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05259128287434578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.052355699241161346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03086814470589161,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02699517272412777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02673065848648548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024733295664191246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024595467373728752,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016258612275123596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016147766262292862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015950508415699005,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010682974942028522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05351909250020981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05351909250020981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.2.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.1790362149477005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.15777480602264404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.14935097098350525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.13222238421440125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.08091238141059875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.07288438826799393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.0961962640285492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.08775460720062256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.083379827439785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.06838435679674149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.06484240293502808,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.04880445823073387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04199089854955673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.03893505781888962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.03818897530436516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.02450370043516159,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.020580368116497993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02021637000143528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.018184015527367592,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.017703499644994736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.013272203505039215,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01354095060378313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.012265851721167564,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.009716490283608437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.04880445823073387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.04880445823073387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.04273678734898567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.03584158048033714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.03184032440185547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.028262462466955185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.01939507946372032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.015999650582671165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.025632822886109352,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.02331860177218914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.020244425162672997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.01563546620309353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.015160813927650452,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.013055362738668919,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.011152705177664757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.009398764930665493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.008941682986915112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.006538650952279568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.004987573716789484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.004653621930629015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.004263903480023146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.003960269037634134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.003473845310509205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0034712685737758875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.002868041628971696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0023902307730168104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.04273678734898567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.04273678734898567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0428481288254261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.03501201048493385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.029858678579330444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.026616744697093964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.019089410081505775,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.01491191703826189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.02664991095662117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.024140529334545135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.020131420344114304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.015170253813266754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.014985787682235241,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.01351439580321312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.011503629386425018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.00922862533479929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.008605791255831718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.006766557227820158,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0048356736078858376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0043977960012853146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.004073855467140675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0036462252028286457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0035126814618706703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0034059712197631598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0026915359776467085,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.002188342157751322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0428481288254261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0428481288254261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1756085604429245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.14955584704875946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.13890667259693146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.12239907681941986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08067318052053452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.06992373615503311,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09721731394529343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08924295753240585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0834602639079094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.06486669182777405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06175591051578522,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04936389625072479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04259105026721954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.03869667276740074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03773403912782669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02467101626098156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01980268396437168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.018941247835755348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.016586411744356155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.015916477888822556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012710344977676868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012334806844592094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011317485012114048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007636150810867548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04936389625072479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04936389625072479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.17265617847442627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.15349774062633514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.14630544185638428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.125789612531662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08070380240678787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07379523664712906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09464309364557266,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0860721543431282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08211905509233475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0660165399312973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06170710548758507,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0487380288541317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.041771307587623596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.039406705647706985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.038827069103717804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024621980264782906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021481908857822418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02098594233393669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018773801624774933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018404876813292503,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013682577759027481,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014801764860749245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012944573536515236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011539943516254425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0487380288541317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0487380288541317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.209043487906456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.19700421392917633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.19321338832378387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.17542071640491486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09851204603910446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09449568390846252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1090313121676445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10056618601083755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0993163213133812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08842168003320694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08447795361280441,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05569034069776535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04819665849208832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04729839414358139,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04708410054445267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02784634381532669,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.024610033258795738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.024359652772545815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02261015959084034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.022482460364699364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.014900386333465576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015216012485325336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.014612419530749321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010649471543729305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04819665849208832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04819665849208832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24484506249427795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23094336688518524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22673256695270538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.2060040980577469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11548089236021042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11090388149023056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1278538554906845,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11774861067533493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11639799177646637,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1037989929318428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09926366060972214,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0653105154633522,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05634693056344986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05533489212393761,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05509016662836075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032666780054569244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028497371822595596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.028219513595104218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026142287999391556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025994718074798584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0174561757594347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017186559736728668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01712212525308132,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011524848639965057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032666780054569244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032666780054569244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.3.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20301590859889984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.17993180453777313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17123068869113922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15232975780963898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09230129420757294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08366638422012329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.10861063003540039,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.09921830892562866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09496082365512848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07847336679697037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07448621094226837,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05513899773359299,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04751267284154892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.044402044266462326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04363902285695076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.027674056589603424,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.023362280800938606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.022996846586465836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.020694028586149216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020208235830068588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.014936521649360657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015182916074991226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.013920888304710388,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.010784035548567772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04751267284154892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04751267284154892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.05948399379849434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05045810714364052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0450628362596035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04024699702858925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.02709878981113434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.022555913776159286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.03599774092435837,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03251658007502556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.02822723612189293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.022175561636686325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02156623639166355,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.01832883432507515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.015611917711794376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.013163863681256771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.012529414147138596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.009210369549691677,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.007052582688629627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.006607340648770332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.006121970247477293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.00571369007229805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.004913266748189926,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.004976646043360233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0040607331320643425,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003532960545271635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05045810714364052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05045810714364052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.05663783848285675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04680328816175461,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04078899696469307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.036554399877786636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.025308020412921906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.020324628800153732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03483179956674576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.031312715262174606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.026685133576393127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02047596499323845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.020128708332777023,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.01776907965540886,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.014995941892266273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01226199883967638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.011548623442649841,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.008920381776988506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.006504755932837725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0060025774873793125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.005570911802351475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005091105587780476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.004694020375609398,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0045917523093521595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0037039704620838165,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.003107720520347357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04680328816175461,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04680328816175461,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1917179524898529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16577664017677307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15542399883270264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1376524418592453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08840838074684143,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07796114683151245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1050894558429718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09659653156995773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09124546498060226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07238545268774033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06868880242109299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05346288904547691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.046208061277866364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04242615029215813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.041495393961668015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.026723310351371765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.021805057302117348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02098340354859829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01855577528476715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0179296862334013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013873790390789509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013583481311798096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012519973330199718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008629865944385529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05346288904547691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05346288904547691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.16939057409763336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1513446867465973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.14464770257472992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.12346041202545166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.079004667699337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07215382903814316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09365367144346237,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.08410616219043732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08046706020832062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06545904278755188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0601811446249485,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04818735271692276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.040927812457084656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.03869343549013138,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.038167115300893784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024444689974188805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021173864603042603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.020702725276350975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018659424036741257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018337540328502655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013722981326282024,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01465285662561655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013013369403779507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011514890007674694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04818735271692276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04818735271692276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1883140653371811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.17655447125434875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1725977212190628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.15641066431999207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0888027548789978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.08470439165830612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09940768033266068,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0913398340344429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08968131244182587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07924304902553558,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07571203261613846,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05076218023896217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04382229968905449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.042683910578489304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.042410291731357574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.025482339784502983,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02221999131143093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02194603905081749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.020312026143074036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.020150791853666306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.013768038712441921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013788512907922268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.013400079682469368,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009614584967494011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05076218023896217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05076218023896217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.246357262134552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23156091570854187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22673998773097992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20575161278247833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11632581055164337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11125487089157104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12949900329113007,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11917395889759064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11740680038928986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10403173416852951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09944429248571396,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06603679060935974,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05707656964659691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05578989163041115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05549166351556778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03308970108628273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028737511485815048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02841159887611866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026251301169395447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02606653794646263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0176161490380764,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01735866442322731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.017196768894791603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011594372801482677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03308970108628273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03308970108628273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.4.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2095576524734497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18695861101150513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17822471261024475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15851232409477234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09566449373960495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08714070916175842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11320783942937851,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10278883576393127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09833946079015732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08187927305698395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0775674432516098,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.0575646311044693,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04935459420084953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046164318919181824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04538174346089363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028994236141443253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024500075727701187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024127205833792686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021858954802155495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02137257158756256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015817200765013695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016173504292964935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014756884425878525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011794302612543106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04935459420084953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04935459420084953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07179942727088928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.06272782385349274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0575268529355526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05125272274017334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03292565792798996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.028575770556926727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04211081191897392,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03832326829433441,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03401795029640198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.027597596868872643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02656625211238861,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.021354198455810547,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01834231987595558,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01593482308089733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.015323616564273834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010706116445362568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008427944034337997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007987068966031075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007394589949399233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.006993317045271397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.005677962210029364,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005734641570597887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004852258134633303,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003964860457926989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05125272274017334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05125272274017334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.06473594158887863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.055526718497276306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.049911461770534515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.044354259967803955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.02931882254779339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.02469693124294281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03860938921570778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03517904505133629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.030482374131679535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.024252815172076225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023485815152525902,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019544633105397224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016757169738411903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.014120960608124733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01342977024614811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009770752862095833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0073308334685862064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006842531729489565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006326897535473108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005861466750502586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0050523970276117325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004929880145937204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004123569931834936,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.003137608990073204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.049911461770534515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.049911461770534515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.20667532086372375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.18412651121616364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.17575180530548096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1560567021369934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09606868773698807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08750919252634048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.11123546212911606,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10240286588668823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09829515218734741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08090247958898544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07650333642959595,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.056504372507333755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04887129366397858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.046007949858903885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.045349981635808945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0281996950507164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.023354772478342056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.022706395015120506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02021184004843235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.019752709195017815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014485612511634827,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013975773938000202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013443267904222012,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008471165783703327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04887129366397858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04887129366397858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18959152698516846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17195124924182892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1653534173965454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14394891262054443,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08860822021961212,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0817338302731514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10249451547861099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0936884954571724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09002437442541122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0746944472193718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07026924192905426,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05280518904328346,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.045185159891843796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04280707612633705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04224920645356178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026574712246656418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02261975221335888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022079620510339737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.019902443513274193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.019540734589099884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01446588896214962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014702496118843555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013650595210492611,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010650047101080418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05280518904328346,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05280518904328346,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1762571483850479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1653290092945099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.16162365674972534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14647088944911957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08305193483829498,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07925103604793549,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09263620525598526,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08550563454627991,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08388134837150574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07411837577819824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07074187695980072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.047173138707876205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.040913499891757965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.039821311831474304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03956005722284317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02358744852244854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020505355671048164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020246082916855812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018697958439588547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018533287569880486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012398071587085724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012418893165886402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012031331658363342,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00827084295451641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.047173138707876205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.047173138707876205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2419440895318985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2276521474123001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2230045050382614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20235870778560638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1143060028553009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10936648398637772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12692731618881226,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11716075241565704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11531758308410645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10232201963663101,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09775715321302414,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06462647765874863,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.056016407907009125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05473296344280243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05443199723958969,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032305825501680374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02803993970155716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027713479474186897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02559611015021801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025405537337064743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016914283856749535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016712086275219917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01648643985390663,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010864359326660633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032305825501680374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032305825501680374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.5.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2036168873310089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1832730621099472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17470474541187286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1561361700296402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0934390053153038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08529166877269745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.111430324614048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10100080072879791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09583073854446411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08062601834535599,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07705391943454742,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05686961114406586,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04845084622502327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04500170052051544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04415525868535042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028722483664751053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02372308075428009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023303192108869553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021242249757051468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020712487399578094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015683121979236603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01551529485732317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014488616958260536,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011047808453440666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04845084622502327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04845084622502327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.06634723395109177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05762307345867157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.051811300218105316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04625110328197479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.030352430418133736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.025681382045149803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04047377035021782,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03665715083479881,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03144051507115364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.025375576689839363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02464967966079712,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.020561281591653824,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.017536645755171776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.014713982120156288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.013981547206640244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010300645604729652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.007822458632290363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007315567694604397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.006863356567919254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.006390023976564407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0054395729675889015,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005470580421388149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004440734162926674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003781738691031933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.051811300218105316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.051811300218105316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.060605134814977646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.051686182618141174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04550164192914963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.04061912000179291,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.02736980840563774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.022464901208877563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03753753378987312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.034015703946352005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.02854038216173649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.022614534944295883,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.022116539999842644,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019017454236745834,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016205081716179848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013237089850008488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.012449914589524269,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009491239674389362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.006952513940632343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006399805191904306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006020466797053814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005488865077495575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0049445740878582,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004851828329265118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.003889617044478655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0031672290060669184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.051686182618141174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.051686182618141174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.18390421569347382,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16390341520309448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1555459350347519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.13815124332904816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0854179784655571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07736478745937347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10099665075540543,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09283767640590668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08745920658111572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07208962738513947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06840716302394867,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.051350172609090805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04431234672665596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04098079353570938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.040139734745025635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02561933919787407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020926428958773613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.020221518352627754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01816825568675995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017614122480154037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013170411810278893,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012800670228898525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011958121322095394,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007893593981862068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.051350172609090805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.051350172609090805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18748553097248077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1659504473209381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15739239752292633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.13692741096019745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08746771514415741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07871513068675995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10398761928081512,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09420737624168396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0894230529665947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07223328202962875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06758613139390945,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.053499579429626465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.045534636825323105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04249802604317665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.041758015751838684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026858501136302948,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02269449643790722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02203347347676754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01975790224969387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.019285235553979874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0145133500918746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015179971233010292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01350901834666729,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011253130622208118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.053499579429626465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.053499579429626465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17114149034023285,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.16083428263664246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15720336139202118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14284665882587433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08073896169662476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07706644386053085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09042980521917343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08328168839216232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08152130991220474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07229864597320557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06913543492555618,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046180956065654755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039897676557302475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03875955566763878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03848888725042343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02312663570046425,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020065199583768845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019804779440164566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018367337062954903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018200399354100227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012346301227807999,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012314205057919025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011960572563111782,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008389642462134361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046180956065654755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046180956065654755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23557148873806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22183120250701904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21720033884048462,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19756180047988892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11134614795446396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10655659437179565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12412071228027344,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11433691531419754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11234697699546814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09990664571523666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0955205112695694,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06329933553934097,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054706476628780365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05334842950105667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05303625017404556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03165601193904877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02740606665611267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027081916108727455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02507716789841652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024877533316612244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016712261363863945,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016471730545163155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016263943165540695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010853748768568039,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05334842950105667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05334842950105667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.6.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21076911687850952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19050660729408264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18252255022525787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16333793103694916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09705882519483566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08929018676280975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11463408172130585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10375270992517471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09938328713178635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08413700014352798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08029516786336899,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05836555361747742,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04982609674334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04680825024843216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.046071410179138184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029368113726377487,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024785494431853294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024426598101854324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02229994349181652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02183464542031288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015965718775987625,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016251832246780396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014962779358029366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011789047159254551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04982609674334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04982609674334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07253632694482803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.06478405743837357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05837472155690193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05205954238772392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03348775953054428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.02869037352502346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04561757668852806,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04102073982357979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03436758741736412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.028703801333904266,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.027966883033514023,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.023239726200699806,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01973414234817028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01628398336470127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.015371035784482956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.011651421897113323,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008739606477320194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.008151110261678696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007829796522855759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.007258750032633543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.006187691818922758,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.006257256492972374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004955670330673456,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.004403706174343824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05205954238772392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05205954238772392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0626121312379837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.05524720624089241,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04855867102742195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.043166615068912506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.028583906590938568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.023709077388048172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04002339765429497,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03640666976571083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.029457036405801773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02430696412920952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023732928559184074,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02031564898788929,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.017455879598855972,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013826129958033562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.012836582958698273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.010184375569224358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.007291768211871386,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006658272352069616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006462015211582184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0058193085715174675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.005293699912726879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.005209335591644049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.00403254572302103,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0033859771210700274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04855867102742195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04855867102742195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19899427890777588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17977093160152435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1722375452518463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.15291020274162292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09298620373010635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0854603573679924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10868975520133972,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09914804995059967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09474433958530426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07913115620613098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07504456490278244,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.055301252752542496,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04738381505012512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04455778747797012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04388604313135147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027598528191447258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022764069959521294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0221426859498024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019917359575629234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.019456520676612854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014231402426958084,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01377798430621624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013111609034240246,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008610040880739689,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04738381505012512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04738381505012512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1884087473154068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.16933833062648773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16095305979251862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1393108069896698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08832570910453796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08029365539550781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10636451095342636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09553088247776031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09012182801961899,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07366209477186203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0699334368109703,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.054981034249067307,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.046462249010801315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04323485121130943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04246465116739273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027733784168958664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023538334295153618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022891977801918983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.020789284259080887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020309122279286385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015384198166429996,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01624443754553795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.014347760006785393,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012545468285679817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.046462249010801315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.046462249010801315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1666836440563202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1567574441432953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1534167230129242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13940130174160004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0788261890411377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07530619204044342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08801373839378357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08102373778820038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07955755293369293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07058826088905334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06748536229133606,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04497779533267021,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03889338672161102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0379004031419754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03766704350709915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02252550795674324,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01975112594664097,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019513443112373352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01810702309012413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0179611723870039,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012069562450051308,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012280250899493694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01174020767211914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008592424914240837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04497779533267021,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04497779533267021,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23579122126102448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.222219780087471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21776892244815826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1981118619441986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1117170974612236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10700331628322601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12449662387371063,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11445385217666626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11269330978393555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10021616518497467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09587167948484421,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06365256011486053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05487082898616791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053646303713321686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.053360287100076675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03187926858663559,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027765214443206787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0274557713419199,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025439215824007988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02525932528078556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01705533266067505,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016955142840743065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016653813421726227,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.0115723367780447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053646303713321686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053646303713321686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.7.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20433014631271362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18405292928218842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1758449673652649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15770600736141205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09402231127023697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08595852553844452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11139664053916931,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10105358064174652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09641910344362259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08123823255300522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07780414819717407,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.056854426860809326,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.048471637070178986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04529405012726784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044517822563648224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028560878708958626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02386007271707058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023480109870433807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021384425461292267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020891698077321053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015477100387215614,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015538507141172886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014423436485230923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011068885214626789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.048471637070178986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.048471637070178986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07229823619127274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0640784278512001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05915708839893341,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05233281850814819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03329877555370331,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.029231058433651924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.042107172310352325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.038525331765413284,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03417966887354851,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.028040915727615356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.026857197284698486,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0213544350117445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01839970052242279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01605963334441185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.01547007542103529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010699591599404812,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008419157937169075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007991536520421505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007386090699583292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.00699745723977685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.005629621911793947,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005631897132843733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0048264265060424805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0037945793010294437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05233281850814819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05233281850814819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.06344876438379288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.05514335259795189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.05007714778184891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.044085338711738586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.028834931552410126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.024654246866703033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03750180825591087,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.034161586314439774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.029770636931061745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.023919442668557167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023006537929177284,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019016969949007034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016308337450027466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013897260650992393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.013273438438773155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009519466198980808,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0072477711364626884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006803502328693867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006268806755542755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0058531141839921474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0049530318938195705,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004873728379607201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004104185849428177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0031792030204087496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.05007714778184891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.05007714778184891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.18240326642990112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1626574695110321,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15496958792209625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.13663387298583984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08455818891525269,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07684142887592316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09996659308671951,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09083674848079681,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08648092299699783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07106637209653854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06717733293771744,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05087890848517418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04350433498620987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04059627652168274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.039900291711091995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.025416448712348938,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020873580127954483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.020246481522917747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.018127888441085815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017663544043898582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013154354877769947,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012906396761536598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012026366777718067,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008315572515130043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05087890848517418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05087890848517418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1977970451116562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18005962669849396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17297670245170593,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15205296874046326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09264685213565826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0856722965836525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10854403674602509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09851948916912079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09423834830522537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07904263585805893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07453982532024384,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05574023351073265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04742798954248428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.044760387390851974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.044106315821409225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027905823662877083,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023507753387093544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022936217486858368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02082017809152603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020411867648363113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014840735122561455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015165221877396107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013953466899693012,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010774858295917511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04742798954248428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04742798954248428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16268427670001984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1532304435968399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14993759989738464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1363285779953003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07694627344608307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0735417902469635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08607260882854462,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07920283079147339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07766101509332657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06899941712617874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0660800039768219,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0439998134970665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.037978317588567734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.036970168352127075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03672965243458748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02201676554977894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01919376291334629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018961550667881966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01760033145546913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017453495413064957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011735353618860245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011853777803480625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011399183422327042,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008189890533685684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0439998134970665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0439998134970665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23203247785568237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21882356703281403,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2143489271402359,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1949910670518875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10994546860456467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10524184256792068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1226196438074112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11273852735757828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11088178306818008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09857067465782166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09438752382993698,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06263895332813263,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05398797243833542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.052726712077856064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.052423760294914246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03131991624832153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027149271219968796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026830697432160378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024843666702508926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02466162107884884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016555219888687134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016400840133428574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01612219400703907,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010934562422335148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05398797243833542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05398797243833542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.8.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21018268167972565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19030331075191498,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18263329565525055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16296975314617157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09686324745416641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0892181545495987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11363700777292252,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10320858657360077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09913817793130875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08380107581615448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0797918289899826,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05793079733848572,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049579475075006485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046735674142837524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04604290425777435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029255900532007217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024781428277492523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02444753423333168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022280484437942505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021841231733560562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016100801527500153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016270168125629425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015168504789471626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.01186355110257864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049579475075006485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049579475075006485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.0923733115196228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08220721781253815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.07622025161981583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.06749071180820465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04258981719613075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.037622444331645966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05377095192670822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04885927587747574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04363660514354706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0360131561756134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03449761122465134,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02736286073923111,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.023438431322574615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02063322439789772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.01992359384894371,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.013733101077377796,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.010986043140292168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010486569255590439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.009727832861244678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009283306077122688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.007337464485317469,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007517325226217508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00634734658524394,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00537061644718051,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05377095192670822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05377095192670822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.07935116440057755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.06979311257600784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06384820491075516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05638270825147629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03625747188925743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.031399913132190704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04671098664402962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04258308187127113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0373074896633625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.030413977801799774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.02923060953617096,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.023759830743074417,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020337557420134544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.017467139288783073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01673617959022522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.011891154572367668,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009111745283007622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008590858429670334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.007955510169267654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007472165394574404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00619489885866642,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006100549828261137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0051757474429905415,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00400660140439868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04671098664402962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04671098664402962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19220881164073944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17436790466308594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.16780315339565277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14817847311496735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0897936075925827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08307909220457077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10316525399684906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09467410296201706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09124304354190826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07631740719079971,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07175160944461823,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.052423689514398575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.045171648263931274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04294484108686447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04242084175348282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.026141708716750145,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.021819069981575012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02130700834095478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019036483019590378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01867504231631756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013404963538050652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012975042685866356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012573492713272572,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007934489287436008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.052423689514398575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.052423689514398575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1993076205253601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18014933168888092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17243245244026184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15134650468826294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09344325959682465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08576633036136627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10941915959119797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09956937283277512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09505053609609604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07860640436410904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07455437630414963,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05619671940803528,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04785337299108505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04504535347223282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04437866061925888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.028088334947824478,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02348434552550316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022888371720910072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02051377296447754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020069656893610954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014821244403719902,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014906519092619419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013882310129702091,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01027657650411129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04785337299108505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04785337299108505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16206996142864227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15213730931282043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14868398010730743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13507655262947083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07658286392688751,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07301879674196243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08578245341777802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07891295105218887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07731891423463821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0684279277920723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06548789143562317,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.043890975415706635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03789806365966797,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.036841198801994324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03659507632255554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021988868713378906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019250184297561646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019006384536623955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017626376822590828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017475441098213196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0117854755371809,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01205415092408657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011435518972575665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008514638058841228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.043890975415706635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.043890975415706635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22811272740364075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21440167725086212,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2097819298505783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19063790142536163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10790391266345978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10301275551319122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12053030729293823,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11082379519939423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10892365127801895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.096433125436306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09228646010160446,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06161956861615181,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053128574043512344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.051801811903715134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05148930847644806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030858976766467094,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026810333132743835,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026479406282305717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024499699473381042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024301273748278618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016448548063635826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016406487673521042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016002364456653595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011173042468726635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053128574043512344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053128574043512344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.9.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21688273549079895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.195871502161026,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18770845234394073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16686895489692688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09991750866174698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09185732901096344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11755452305078506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10667914897203445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10227162390947342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08605503290891647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0815466120839119,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05971937254071236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05112798139452934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.048075463622808456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04733084887266159,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030013510957360268,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02524554915726185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024882636964321136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022554337978363037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022082101553678513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016292694956064224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016276968643069267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015285375528037548,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011505048722028732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05112798139452934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05112798139452934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.08111102879047394,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.07254783809185028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.06766746938228607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05942074581980705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.037567462772130966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.03346123546361923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04657381772994995,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.042521990835666656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03834857791662216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03160262852907181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.030077604576945305,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.023668518289923668,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.020309679210186005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.018092622980475426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.017540713772177696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.011844425462186337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.00947027001529932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.009059125557541847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.008297783322632313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.007934422232210636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.006231253035366535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.006245048716664314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00544535368680954,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.004236916080117226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04657381772994995,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04657381772994995,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.07359469681978226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.06516092270612717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06042276322841644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05277974158525467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.033782899379730225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.029807444661855698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.042435433715581894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03851376846432686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.034619253128767014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02816605381667614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.026803895831108093,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02149576134979725,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.018389325588941574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.016235601156949997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01569240540266037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.010735702700912952,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.008401831611990929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.007987060584127903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.007274224888533354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.00690606702119112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0055682044476270676,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.005442801862955093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004773732740432024,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0034874265547841787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05277974158525467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05277974158525467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.17976276576519012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16189919412136078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15510153770446777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1357804536819458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08386819809675217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07699833065271378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09768102318048477,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08885928243398666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08545619249343872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07042080163955688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06619781255722046,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.049674712121486664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04243537038564682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04017745330929756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03963658586144447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.024822253733873367,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020555876195430756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02002415992319584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.017803162336349487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017433637753129005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012822971679270267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012464363127946854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011917990632355213,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007906206883490086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.049674712121486664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.049674712121486664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19702740013599396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17771965265274048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16960251331329346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1487310379743576,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09191865473985672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08419815450906754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10898296535015106,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09889288991689682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09361158311367035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0775689110159874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07357574999332428,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.056102555245161057,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04769488424062729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04453796520829201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04377840459346771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.028172001242637634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023658720776438713,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02303505875170231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0208913441747427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020396705716848373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015303699299693108,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01567230373620987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01426894124597311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01145181804895401,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04769488424062729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04769488424062729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16228733956813812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15147458016872406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.147598534822464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.133662149310112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07651983946561813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07260174304246902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08628086000680923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07926963269710541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07741113752126694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06795590370893478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0649128332734108,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04409165307879448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03805132955312729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03680633008480072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03651457279920578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022104473784565926,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019228477030992508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018950073048472404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01753089763224125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0173508133739233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011845353990793228,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012083268724381924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011417168192565441,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00851461011916399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04409165307879448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04409165307879448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22261309623718262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.20802195370197296,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20303837954998016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18380849063396454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1050167977809906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0998120978474617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11781039834022522,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10840636491775513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10619650781154633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0932191014289856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0889882817864418,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06010079383850098,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0518934428691864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.050366226583719254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.050000064074993134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030053550377488136,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.025963500142097473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02559644542634487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02357950061559677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023345347493886948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.015806002542376518,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015786699950695038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015273896045982838,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010550056584179401,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0518934428691864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0518934428691864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.10.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22963909804821014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20756880939006805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19922778010368347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1766587495803833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1061212494969368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0978289544582367,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12449796497821808,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11283789575099945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10863349586725235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09140748530626297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08657213300466537,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06365633755922318,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0542290173470974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.051196929067373276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05046088993549347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03204844519495964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027043066918849945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026687270030379295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02419471926987171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023724334314465523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017471883445978165,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017611542716622353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01644137129187584,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012711179442703724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0542290173470974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0542290173470974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09471038728952408,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08441315591335297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.07877811789512634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.06901639699935913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04399825632572174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.03916407749056816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05446501821279526,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04960593208670616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04493807256221771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03683320805430412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03501424565911293,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02770739234983921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.023780781775712967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.021279659122228622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.020652614533901215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.013886299915611744,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.011258203536272049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010791515931487083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.009872546419501305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009467078372836113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0073661478236317635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007558855228126049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006490133237093687,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005331674125045538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04960593208670616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04960593208670616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0793822705745697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07023479789495468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06453816592693329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05635121092200279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03643607348203659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03185543790459633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04654272645711899,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04240185767412186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.03729833662509918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03040153533220291,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.02904202975332737,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.023674670606851578,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020249824970960617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01757160946726799,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.016896599903702736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.011824156157672405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009196280501782894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008705075830221176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00800666119903326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007557692006230354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00619993731379509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006152571178972721,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005265026353299618,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004109062720090151,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04654272645711899,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04654272645711899,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1938401460647583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17283737659454346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1647641360759735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14326342940330505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09012022614479065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08195556700229645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10631053894758224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0960053950548172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09189220517873764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07470855861902237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06997879594564438,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054187141358852386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04590385779738426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04318470135331154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04252967983484268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027100728824734688,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022095059975981712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021475177258253098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01895921491086483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.018507080152630806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013989497907459736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013476541265845299,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012775829993188381,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008517906069755554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054187141358852386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054187141358852386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19575975835323334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17775772511959076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16314177215099335,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14397133886814117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09221747517585754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08105906844139099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12340552359819412,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10978596657514572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0938209667801857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0790930762887001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0766202062368393,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06391671299934387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05313868075609207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04506020247936249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04295142740011215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032097749412059784,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.024306951090693474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022937802597880363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02180151827633381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020510662347078323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017426537349820137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01731007546186447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0146400835365057,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012545672245323658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05313868075609207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05313868075609207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1656651645898819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15426109731197357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15001104772090912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13573534786701202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.078199602663517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07397285103797913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08861343562602997,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08129531145095825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07914450764656067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06920995563268661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06611698120832443,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04535221680998802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03911295160651207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03772561997175217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03739975392818451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022767936810851097,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019898805767297745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01959347166121006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01814555749297142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017941074445843697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012317722663283348,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012772978283464909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011842994019389153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009278111159801483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04535221680998802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04535221680998802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2246875762939453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2094203233718872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20410379767417908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18452906608581543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10587392002344131,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10035202652215958,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1192048043012619,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10953209549188614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1070624440908432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09369143098592758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0893295407295227,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06066050007939339,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05242482200264931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05077127739787102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05038008093833923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030338531360030174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026141522452235222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025747839361429214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02367529831826687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023426810279488564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0158806461840868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015857579186558723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01529465802013874,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010516893118619919,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05242482200264931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05242482200264931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.11.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22915378212928772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20730051398277283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19904541969299316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17664343118667603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10618459433317184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09784796833992004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12465621531009674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1128915399312973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10870573669672012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09153155237436295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08664526045322418,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06376823782920837,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05429444834589958,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05123336985707283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05048857629299164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03208930417895317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027049902826547623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026694169268012047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02420920506119728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023741841316223145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017339862883090973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017597826197743416,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.016272757202386856,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012688130140304565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05123336985707283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05123336985707283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09936142712831497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0882679894566536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08140312135219574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07137010246515274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04610950127243996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04047267511487007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.059583116322755814,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05331159383058548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04718722775578499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03861810266971588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03724299371242523,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.030456092208623886,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.025676630437374115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02238871343433857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.021564612165093422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.015369811095297337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0120000084862113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011409792117774487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010573736391961575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010039178654551506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.008258509449660778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008328622207045555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007047239691019058,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0059978412464261055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05331159383058548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05331159383058548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08464791625738144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07457271218299866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.0683441236615181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05965745449066162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0388471893966198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03373681753873825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.050359148532152176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04534270614385605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.039873044937849045,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03232743963599205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.030980568379163742,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025655295699834824,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0217230673879385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01878642849624157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018025169149041176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012857151217758656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009926670230925083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009392314590513706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008654155768454075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008158053271472454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006775897927582264,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006758794654160738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005715447477996349,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004645415581762791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.050359148532152176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.050359148532152176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19185824692249298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17256228625774384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.16470789909362793,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14446885883808136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08987101912498474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08207979798316956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10619068145751953,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09596749395132065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09148573130369186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07533403486013412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07105202227830887,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054195694625377655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04592197388410568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04309283569455147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04241381585597992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027088407427072525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02205561101436615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021414682269096375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019084369763731956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01862308569252491,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013995630666613579,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01345471478998661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012836312875151634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008503880351781845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054195694625377655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054195694625377655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.20962904393672943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18562078475952148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17620067298412323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15445590019226074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09794670343399048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08820542693138123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.116356261074543,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10519334673881531,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10026030987501144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08127445727586746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07676979154348373,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06004529073834419,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05087818577885628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04757343977689743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04677799716591835,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03023526445031166,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.025310833007097244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.024546710774302483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.022090625017881393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.021575355902314186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01639850251376629,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0168004110455513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01531671267002821,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012325716204941273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05087818577885628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05087818577885628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16898709535598755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1571710705757141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15290850400924683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1381736844778061,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07989440113306046,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07549551129341125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09060564637184143,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08289023488759995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08087866753339767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07055147737264633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06759300827980042,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04641929268836975,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039985645562410355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.038659267127513885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03834491968154907,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02337837591767311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02056017331779003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020253784954547882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018753837794065475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018560145050287247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012866825796663761,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013408723287284374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01244029775261879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009980638511478901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04641929268836975,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04641929268836975,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23236511647701263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21642330288887024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2109871655702591,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1904449611902237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10978308320045471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10397175699472427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12320615351200104,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11323282122612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11103887110948563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09688219428062439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09235727041959763,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06292006373405457,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054283592849969864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.052717555314302444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05233968421816826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03146981820464134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02724529057741165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0268524382263422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024647196754813194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02441091276705265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016622863709926605,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016672734171152115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01608199253678322,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01129305362701416,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054283592849969864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054283592849969864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.12.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.23011666536331177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.2070072591304779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19730734825134277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17449317872524261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1063273698091507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09702161699533463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12708328664302826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11497606337070465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10899266600608826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09107682853937149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08619516342878342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06467711925506592,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05527600273489952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05132497847080231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05035177245736122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03264697268605232,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02715372107923031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02667497657239437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.0242130346596241,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023597614839673042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017836224287748337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017873873934149742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01653091423213482,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012847579084336758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05132497847080231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05132497847080231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10213909298181534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09181688725948334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08669238537549973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07614187151193619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04754956066608429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04299907013773918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.057521212846040726,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05237749591469765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0484396331012249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.040061332285404205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03804078698158264,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.029256463050842285,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02503294311463833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0228714719414711,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022345155477523804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01463894173502922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0118982819840312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011480354703962803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010407092981040478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.01006257627159357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.007670809514820576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00764026353135705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006871089804917574,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005136272870004177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05237749591469765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05237749591469765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08364420384168625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07518991827964783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07015972584486008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06166722998023033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03878398612141609,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03463580459356308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04810282588005066,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.043912373483181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.039543092250823975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.032745540142059326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.031173640862107277,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.024396320804953575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020994562655687332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018659228459000587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018081730231642723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012200326658785343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009689881466329098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0092628113925457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008484462276101112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008097478188574314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006369702983647585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006301886402070522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.00556590873748064,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004146175924688578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04810282588005066,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04810282588005066,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.20933467149734497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1882956326007843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18069924414157867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.158315509557724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09775286912918091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08979347348213196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1138065904378891,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10322064906358719,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09948767721652985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08194328844547272,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07698874920606613,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05798124149441719,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04929068312048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04681167006492615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04621042311191559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.028967253863811493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02381446212530136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02323269098997116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02055460587143898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02015010267496109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014868661761283875,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.014231311157345772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013744518160820007,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008764226920902729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04929068312048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04929068312048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.22750461101531982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.2072417140007019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19912861287593842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17833483219146729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10650797188282013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09856891632080078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12446413189172745,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11342708766460419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10829344391822815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09154056757688522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08819243311882019,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06406870484352112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0546913668513298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05158976837992668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.050834186375141144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032115962356328964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027341658249497414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026722058653831482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.024411490187048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.023940661922097206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01732906885445118,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.017942391335964203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016332853585481644,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013093994930386543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05158976837992668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05158976837992668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17158488929271698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1596268117427826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15523813664913177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14031267166137695,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08117655664682388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07666884362697601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09198509156703949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08441497385501862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08218464255332947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07170307636260986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06856146454811096,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04721067100763321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04066522419452667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.039182402193546295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03883276879787445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02365320362150669,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020666969940066338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020345376804471016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01881992071866989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018607188016176224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012764004059135914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013277272693812847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012260274961590767,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009646973572671413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04721067100763321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04721067100763321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23705653846263885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22088705003261566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2152370661497116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19453756511211395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1120653972029686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10619194060564041,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1260979324579239,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11580414324998856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11338284611701965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09901244193315506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0943223088979721,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06442520767450333,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.055486634373664856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05379866436123848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05340783670544624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032245948910713196,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02780304290354252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02740365080535412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025173520669341087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024921493604779243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017092501744627953,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01700826920568943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016501493752002716,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011473393999040127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05379866436123848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05379866436123848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.13.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2439960241317749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.21875041723251343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.20916424691677094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.18457379937171936,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1127302423119545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.1031016856431961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.13287678360939026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1204565018415451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.11572746932506561,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09608486294746399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0907578244805336,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06803296506404877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05786415562033653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05433708801865578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05347524583339691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03423671796917915,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.0285765640437603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.0281657837331295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02533121034502983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.0247786957770586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.018537364900112152,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01849132962524891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.017353331670165062,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013140566647052765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05347524583339691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05347524583339691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10424488037824631,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09404675662517548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08792061358690262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07763238996267319,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.048578958958387375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.043544284999370575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06060922145843506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0549757182598114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04951776936650276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04127885401248932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03949272260069847,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.030869731679558754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02632269635796547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.023477280512452126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022763244807720184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.015449536964297295,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012360786087810993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011845728382468224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010933089070022106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010474280454218388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00815539713948965,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00824003480374813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0071189384907484055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005706408992409706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04951776936650276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04951776936650276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0841318741440773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07603752613067627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.0695098266005516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.061464838683605194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.038894351571798325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03407863900065422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.051064424216747284,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04636864736676216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.03965947777032852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.033343810588121414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03209308907389641,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025806216523051262,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.022154830396175385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018763301894068718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.017897984012961388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012900080531835556,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009814596734941006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00923085305839777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00871579721570015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008151985704898834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006758652627468109,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00665030675008893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005608730483800173,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004370537586510181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.051064424216747284,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.051064424216747284,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2194114476442337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19817328453063965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1897161304950714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1665915846824646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10261544585227966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09422661364078522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12187033146619797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10919661074876785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10440018773078918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08657117933034897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08166251331567764,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06234155595302582,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05223219096660614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04920317605137825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04848750680685043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0311729833483696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025239553302526474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024584274739027023,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.021982988342642784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021497495472431183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016114113852381706,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015443004667758942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014659930020570755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009926020167768002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05223219096660614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05223219096660614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2451043426990509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.21659700572490692,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.20483578741550446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17874456942081451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.11447533965110779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.10334701091051102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.1379907876253128,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.12453643232584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.11774825304746628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09409315139055252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0903477817773819,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.07104182243347168,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.06020486354827881,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.055550310760736465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.054439641535282135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03559456765651703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.029506895691156387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02864748425781727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.025540543720126152,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02481147274374962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.019040733575820923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.019608929753303528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01755474880337715,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.014199887402355671,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03559456765651703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03559456765651703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1612621545791626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15026816725730896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14586375653743744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13207511603832245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07646261900663376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07212352752685547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08729451894760132,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08010675758123398,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07737758010625839,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06777670234441757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06485375761985779,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044831451028585434,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03866015747189522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037003252655267715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03661582991480827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022499021142721176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019655419513583183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019326208159327507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017969045788049698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01773681491613388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012186964973807335,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01285128016024828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011632228270173073,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009495903737843037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044831451028585434,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044831451028585434,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2333511859178543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2181156873703003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21250203251838684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19235879182815552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11058540642261505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10485655814409256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12495282292366028,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11465955525636673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1117902547121048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09811998158693314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09367595613002777,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06389470398426056,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.055031534284353256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05319531261920929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05276918783783913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0320056714117527,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02765602432191372,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02723834291100502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02515741065144539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024881061166524887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01706412434577942,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017189128324389458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016435060650110245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01187122892588377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05319531261920929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05319531261920929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.14.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22905723750591278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20517916977405548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19521762430667877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1729503870010376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10571648925542831,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09612210839986801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12685558199882507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11451590061187744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.1084742546081543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09032357484102249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08582853525876999,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06501206755638123,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05518336594104767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0513480007648468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.0503263846039772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03308285027742386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02759529836475849,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.027087146416306496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.024699050933122635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.024051770567893982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01840297505259514,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.018716109916567802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01693183183670044,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013982602395117283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0513480007648468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0513480007648468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11666414886713028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10602928698062897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09970837086439133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08854666352272034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05452314764261246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.049299050122499466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06797266751527786,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06117452681064606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.055459439754486084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04683902859687805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04488718509674072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03467332944273949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02940622717142105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.026408664882183075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025664405897259712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0174703486263752,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014032304286956787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013494862243533134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01255026925355196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012077713385224342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009334007278084755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009463821537792683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008200091309845448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006772384513169527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04683902859687805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04683902859687805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08810259401798248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08004835993051529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07284713536500931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06478346884250641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04067221283912659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03553685545921326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05450048670172691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04918598383665085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.041490521281957626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03524171561002731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03419729322195053,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.027716979384422302,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.023541107773780823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.019659943878650665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01862768456339836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.013893683440983295,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010324222035706043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00965107511729002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009246932342648506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008591270074248314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007267958018928766,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007098592352122068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005894139409065247,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004685773514211178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04918598383665085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04918598383665085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23456385731697083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21531420946121216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.20743320882320404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1857202649116516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11070002615451813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10285980999469757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1297006905078888,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1173039972782135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11238881200551987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09601468592882156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09145043045282364,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06627190858125687,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0561567060649395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05312134325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05239717289805412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033112697303295135,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027156980708241463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.026501482352614403,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02409745194017887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023611770942807198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01705765165388584,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01640031486749649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015800446271896362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01031299214810133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05312134325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05312134325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21899835765361786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1978422850370407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19002032279968262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16452579200267792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10246901959180832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09439628571271896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11919937282800674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10794547945261002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10393045097589493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08561493456363678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07929619401693344,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06125788763165474,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0518881231546402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0493588000535965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.048741310834884644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03068169765174389,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02574397437274456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025154128670692444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02237648144364357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.021974915638566017,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016339421272277832,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01626567915081978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015499301254749298,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011294243857264519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0518881231546402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0518881231546402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15770180523395538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14742374420166016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14333368837833405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13005302846431732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.074795663356781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07083833962678909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08524028956890106,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07809461653232574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07567718625068665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06661991029977798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06384902447462082,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04373669996857643,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03773334249854088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03623216971755028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03588201850652695,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022020608186721802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01930447854101658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01900804601609707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017728151753544807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017513221129775047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012091686949133873,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012669697403907776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011600361205637455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00945026334375143,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04373669996857643,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04373669996857643,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23564356565475464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2211351990699768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21591977775096893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19587956368923187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11163628846406937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10627896338701248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12551924586296082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11538129299879074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11277883499860764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09955240041017532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09522084146738052,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06404978036880493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05523828789591789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053552087396383286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05315331742167473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03202453628182411,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027533670887351036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027150021865963936,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025073280557990074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024819286540150642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016862886026501656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016634929925203323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01628117449581623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010960489511489868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053552087396383286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053552087396383286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.15.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22647976875305176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.2024482637643814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1927223652601242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17141802608966827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10423047840595245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0946464091539383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12394973635673523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11255349963903427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.1070900559425354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08901431411504745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08450054377317429,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06320696324110031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05404765158891678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05026623606681824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04934534803032875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.031886156648397446,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02654283121228218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026087269186973572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023602349683642387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023010283708572388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017442844808101654,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017400845885276794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01620781607925892,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012444248422980309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05404765158891678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05404765158891678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11590727418661118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10605204105377197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10048915445804596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08960055559873581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05413498356938362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.049528010189533234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06562458723783493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.059830501675605774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05495759844779968,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04686499014496803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04476265236735344,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0334007665514946,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028617089614272118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.026046190410852432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025423923507332802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016733521595597267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01356329396367073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013093114830553532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012112772092223167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011715425178408623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00880417414009571,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008751925081014633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007870012894272804,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005899733863770962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05413498356938362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05413498356938362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09108318388462067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08345703780651093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07700448483228683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06886369735002518,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04217318817973137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.037533994764089584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.054534588009119034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.049892932176589966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042882587760686874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.036832045763731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.035446710884571075,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02755543403327465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.023793211206793785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.020290987566113472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019393084570765495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.013776133768260479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010540287010371685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00992585439234972,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00948261097073555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008879275992512703,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007188939023762941,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007032295223325491,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005993238650262356,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004482063464820385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.049892932176589966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.049892932176589966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23983266949653625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.22201016545295715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.21503150463104248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1933378279209137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11326155811548233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10614102333784103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13152951002120972,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11893036216497421,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11476211249828339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09913219511508942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09454678744077682,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06720900535583496,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05683869868516922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05423944443464279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05360081046819687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03359147533774376,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027577079832553864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.027025625109672546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.024652138352394104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.024242153391242027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01724315993487835,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01637602411210537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.016067082062363625,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010034640319645405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05423944443464279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05423944443464279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2320849895477295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.20865845680236816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.20059821009635925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1781105250120163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10818787664175034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.099416084587574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12471873313188553,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11383410543203354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.11023689061403275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09156417101621628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08662991970777512,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06421130895614624,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05503349006175995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05245098099112511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05181943252682686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0323946475982666,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027934659272432327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.027320686727762222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.024720456451177597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02433253638446331,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017744649201631546,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01840244047343731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01691754162311554,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013658495619893074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05245098099112511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05245098099112511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14805495738983154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13856996595859528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13454031944274902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12218108773231506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07016174495220184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06640314310789108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08033265918493271,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07361839711666107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0709780603647232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06259211897850037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06004016101360321,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04113738611340523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03550390899181366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03396492078900337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0335921049118042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.020667053759098053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018040912225842476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01773693785071373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.016575664281845093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0163434948772192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011241261847317219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011804170906543732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010727626271545887,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008715817704796791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04113738611340523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04113738611340523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22845053672790527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21470676362514496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20956233143806458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19028256833553314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1082024797797203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.1030900627374649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12179800122976303,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11202165484428406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10928066819906235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09672294557094574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09250153601169586,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06213308870792389,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05364133045077324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0519101582467556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.051498766988515854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.031056690961122513,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026702728122472763,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026321053504943848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024355579167604446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02410231903195381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016307581216096878,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016161657869815826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015705900266766548,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010651995427906513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05364133045077324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05364133045077324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.16.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21916551887989044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1945718377828598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1843707263469696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16384552419185638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1004827544093132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09058739244937897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12038393318653107,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10945025086402893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10355225205421448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08526989817619324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08110543340444565,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06147739291191101,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05259532108902931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04852879047393799,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04750606790184975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03106815181672573,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025667477399110794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02517629787325859,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022724425420165062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022072019055485725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016971994191408157,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01695694774389267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015614386647939682,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012133860029280186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05259532108902931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05259532108902931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09702017903327942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08892075717449188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08326403051614761,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0748206302523613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.045153599232435226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04079575836658478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05668080598115921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0516507513821125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04597777873277664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03949667513370514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03810432553291321,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.028803909197449684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02470206469297409,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.021792028099298477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.021063625812530518,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.014429976232349873,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.011458257213234901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010961330495774746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01034674234688282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009877747856080532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0075792958959937096,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007655511610209942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006535821128636599,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005238090176135302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0516507513821125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0516507513821125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08191414177417755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0746629536151886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06833002716302872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.061438750475645065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03774355724453926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03322683274745941,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05006025731563568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04548916965723038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.038546204566955566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03305013105273247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03211662173271179,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025281812995672226,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02165038511157036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018216287717223167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01730922982096672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012625445611774921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009527072310447693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008940786123275757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008605592884123325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008032230660319328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006586558651179075,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006493933964520693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005386063363403082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0042536910623312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05006025731563568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05006025731563568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19983750581741333,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.18037481606006622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.17052434384822845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.15358620882034302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09235408902168274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08370831608772278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1147618442773819,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10271640121936798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09462103247642517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08016729354858398,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07711547613143921,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.058539628982543945,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04904281347990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0443052276968956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04314000904560089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.029245130717754364,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022769056260585785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021893257275223732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.020283309742808342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01951954886317253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0150186438113451,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01434130035340786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012997607700526714,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009044409729540348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04904281347990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04904281347990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2257997989654541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.20301946997642517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19423991441726685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17116889357566833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10530448704957962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09642394632101059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12401852011680603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.1123625859618187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10746617615222931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08944348245859146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08399651944637299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0638689175248146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054264314472675323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05109007656574249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05034219101071358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032092299312353134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0271917674690485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026480693370103836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02412840910255909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02363264374434948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017424728721380234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.018006546422839165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016407500952482224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013257740996778011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054264314472675323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054264314472675323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15382196009159088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14381498098373413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1396910846233368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12702886760234833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07270950824022293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06881019473075867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08308828622102737,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07620883733034134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07357731461524963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06495369225740433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.062314413487911224,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.042544521391391754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03672734647989273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.035143643617630005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03476990759372711,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021390043199062347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018609371036291122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018298383802175522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017106514424085617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.016874471679329872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011657150462269783,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012097058817744255,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011129972524940968,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008852720260620117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.042544521391391754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.042544521391391754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23629136383533478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.221998929977417,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2167816162109375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1970624476671219,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11181584000587463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10657082498073578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12595754861831665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11568234115839005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1129145696759224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10006839781999588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0960417091846466,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06434372812509537,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05548098310828209,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05371786653995514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05330934748053551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032316654920578,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027896175161004066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027507085353136063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025547264143824577,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025291267782449722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017312465235590935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017253439873456955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016725366935133934,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011848049238324165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05371786653995514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05371786653995514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.17.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.23277296125888824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20573410391807556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1946689784526825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17207078635692596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10690053552389145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09603912383317947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12889841198921204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11631854623556137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.11022789776325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09005008637905121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.085513174533844,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06579466909170151,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05594460293650627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05167718231678009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.050617765635252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.033234138041734695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027429264038801193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026894068345427513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.024185629561543465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023510124534368515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0182279571890831,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.018219564110040665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.0167915690690279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013175844214856625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05167718231678009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05167718231678009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11159697920084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10284176468849182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09691385924816132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08699440956115723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05219624936580658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04757829010486603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06522989273071289,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05899149924516678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05303889140486717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04585108160972595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04416908323764801,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.033286046236753464,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028333332389593124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025280551984906197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02452143095433712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016733825206756592,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013424481265246868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012912550009787083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012190655805170536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011723286472260952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.008938467130064964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009078599512577057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00782180018723011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006471633445471525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05303889140486717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05303889140486717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09080873429775238,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0825132355093956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.074677474796772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06732600182294846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04180391505360603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03631275147199631,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05631629005074501,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0514843687415123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042789045721292496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03666194528341293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03579147905111313,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02847900614142418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.024534741416573524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02019139565527439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019041620194911957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014246593229472637,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010592938400804996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00984676368534565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009584269486367702,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008834848180413246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007452117744833231,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007347347680479288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005969054065644741,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004772136453539133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0514843687415123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0514843687415123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.21817098557949066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19805368781089783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18795980513095856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17003194987773895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10196997970342636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09285135567188263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12577274441719055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1121627613902092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10425809025764465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08868373185396194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08565255254507065,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0645514577627182,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053657468408346176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.048984818160533905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04783867299556732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03234352543950081,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02512260526418686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024219023063778877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.022392522543668747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021634038537740707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016693469136953354,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015652792528271675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01456428226083517,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009827886708080769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053657468408346176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053657468408346176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19611601531505585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17451535165309906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16691315174102783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14427968859672546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0910203754901886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08288411796092987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10581089556217194,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09621437638998032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09244778007268906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07538393139839172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07042215764522552,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05457882955670357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04667678102850914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04434116557240486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04377315938472748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027627406641840935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02394750341773033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.023392587900161743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02099853940308094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020642446354031563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015439669601619244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016207506880164146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01468756515532732,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012435068376362324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04667678102850914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04667678102850914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14265544712543488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13327494263648987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12921646237373352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.11749128997325897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06741973012685776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06365444511175156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07725055515766144,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07099489867687225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06826197355985641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.060155946761369705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05766630917787552,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03952312469482422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0341607928276062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.032549675554037094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03216953203082085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019820544868707657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017160383984446526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01685040071606636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015736378729343414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015500363893806934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010679199360311031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011069446802139282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01013621874153614,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007981355302035809,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03952312469482422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03952312469482422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22432978451251984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21059300005435944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20538759231567383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.186675027012825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10603143274784088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10088574141263962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11982519179582596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.1101493239402771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10717980563640594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09490736573934555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09087194502353668,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061201393604278564,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05278884619474411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05094105750322342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.050502434372901917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030659247189760208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026339607313275337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02594776079058647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02409440465271473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023816602304577827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016307314857840538,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01617657206952572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015671780332922935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010914131067693233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05278884619474411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05278884619474411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.18.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22135470807552338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19446787238121033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1830441653728485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16147533059120178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10138344764709473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09030714631080627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12372703105211258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11153721064329147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10473138839006424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08512913435697556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08101708441972733,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06333307176828384,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05381900072097778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0491536520421505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04799271002411842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03215419501066208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026349464431405067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025791587308049202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023273451253771782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022537147626280785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017878437414765358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01791355572640896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.016312770545482635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013245079666376114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05381900072097778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05381900072097778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11465656012296677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10600828379392624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09989749640226364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0902431458234787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053645115345716476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0489417165517807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06711427122354507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.060852937400341034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05449691414833069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04747826233506203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.045961130410432816,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03420788794755936,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.029157904908061028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025898372754454613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025087988004088402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017144372686743736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013635821640491486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013091490603983402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012428264133632183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011927100829780102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00912641640752554,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009099568240344524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007961972616612911,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006304722744971514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053645115345716476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053645115345716476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08965000510215759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08232685178518295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07426565140485764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06724289059638977,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04138024151325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0359480194747448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.056460265070199966,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.051676955074071884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042223282158374786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0367119237780571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03599132224917412,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02862909622490406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02458825148642063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02000809647142887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018770568072795868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014324580319225788,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010489951819181442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009705272503197193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009572271257638931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008795415051281452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007511377800256014,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0073248897679150105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005949682090431452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004749493673443794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.051676955074071884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.051676955074071884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22832971811294556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2091153860092163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.20005400478839874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.18111038208007812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1068718433380127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09854882955551147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12875410914421082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11564058065414429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10896963626146317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09360790997743607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08994371443986893,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06586138904094696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05534180998802185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123921111226082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.050247516483068466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.032931022346019745,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026108600199222565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025332389399409294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023362506181001663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022718077525496483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01687576062977314,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015892764553427696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015032531693577766,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00970155093818903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123921111226082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123921111226082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21137022972106934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1925043761730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.18610072135925293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1632358729839325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09859667718410492,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09163976460695267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11444369703531265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10292671620845795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09996694326400757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08387401700019836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0796528160572052,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.059008631855249405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05010414123535156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04815682768821716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.047679923474788666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02986222133040428,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.026172997429966927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025721712037920952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02343345619738102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.0231548510491848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016754725947976112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01785171777009964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016134297475218773,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01397703867405653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05010414123535156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05010414123535156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.13911974430084229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.130160853266716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12640529870986938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.11501024663448334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06569506973028183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06214374676346779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07496024668216705,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.06896717846393585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06651759892702103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.05869591608643532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05628379061818123,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03827868029475212,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.033143602311611176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03168552368879318,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03133014962077141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019200792536139488,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.016632873564958572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01634667068719864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015251516364514828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015037299133837223,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010289926081895828,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010616136714816093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.009803997352719307,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007561687845736742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03827868029475212,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03827868029475212,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.21969464421272278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.20638231933116913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20135000348091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18314434587955475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1038883775472641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09891800582408905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11729646474123001,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10784762352705002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10504719614982605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09307779371738434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.089162178337574,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05984938517212868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05170333757996559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04993525519967079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04951624944806099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030045464634895325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02594025991857052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025562485679984093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02376994490623474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02350855991244316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016147594898939133,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01608770340681076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015559244900941849,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011065609753131866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05170333757996559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05170333757996559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.19.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20882654190063477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18293501436710358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1708936095237732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1502760797739029,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09569812566041946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08449085056781769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11814698576927185,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10681642591953278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09894713014364243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07988262176513672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0761018916964531,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06059737503528595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05178295448422432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046530455350875854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04524950683116913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03089858964085579,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025215450674295425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02453891932964325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02227173186838627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02141343243420124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017463907599449158,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017549382522702217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015780964866280556,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013132983818650246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05178295448422432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05178295448422432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11444326490163803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1055014580488205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09825535863637924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0891423374414444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053389087319374084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04801574721932411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06906714290380478,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06236892566084862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05436888337135315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0473749116063118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04616570472717285,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03520072624087334,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02987632527947426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0258669164031744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024836190044879913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01766885630786419,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01373202446848154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013081732206046581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012574264779686928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011948813684284687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009422007016837597,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009440392255783081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007976274937391281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006634380668401718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053389087319374084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053389087319374084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09047260880470276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08232685923576355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07258374243974686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06598429381847382,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04149811714887619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.035029057413339615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0590379424393177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053789906203746796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04254136607050896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0367581807076931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.036301515996456146,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.029818270355463028,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.025577928870916367,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.020057296380400658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018573004752397537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014916702173650265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010609252378344536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009675871580839157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009685897268354893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008738551288843155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00780309597030282,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00764507194980979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005905516445636749,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004975953139364719,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053789906203746796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053789906203746796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22678525745868683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2049216777086258,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19451342523097992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1758013367652893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10528506338596344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09566793590784073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12972676753997803,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11574698984622955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10772284865379333,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09136522561311722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08834898471832275,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06667500734329224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05533298850059509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05051372945308685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049303170293569565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03336772322654724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025823956355452538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024896109476685524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02298101596534252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02219715341925621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0171296838670969,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016026947647333145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014846895821392536,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00993427261710167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05051372945308685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05051372945308685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.22984875738620758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.2021685689687729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19181154668331146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16082192957401276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10723299533128738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0965796411037445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12582704424858093,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11445076018571854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10925163328647614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08504299819469452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.079674132168293,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06471829116344452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05526352301239967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.051943499594926834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05112643912434578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032539453357458115,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027630910277366638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02689254656434059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.023370549082756042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.022835325449705124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017570000141859055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.018266256898641586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016515854746103287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013427401892840862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.051943499594926834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.051943499594926834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1456148475408554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1367599219083786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13323040306568146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12116726487874985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06882210820913315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06545256078243256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07771433144807816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07165767252445221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06958616524934769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06164563074707985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05905027315020561,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03964614123106003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03439237177371979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.033128201961517334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.032827407121658325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019848518073558807,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017282966524362564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.017022754997015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015853779390454292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015668600797653198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010559497401118279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010853266343474388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01013503223657608,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007590027526021004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03964614123106003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03964614123106003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22867681086063385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21522530913352966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2104169726371765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19131822884082794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10803673416376114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10315576940774918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12112215906381607,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11150986701250076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10908965766429901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09691968560218811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09275484830141068,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061633262783288956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05334612727165222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05178083851933479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05141318216919899,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030911065638065338,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026616882532835007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026269732043147087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024352600798010826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02411971427500248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01639235019683838,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016056003049016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015872187912464142,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010563229210674763,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05334612727165222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05334612727165222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.20.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20761169493198395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18257753551006317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17165516316890717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15128041803836823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09508275985717773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08455172181129456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11502408981323242,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10470683872699738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09812071174383163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07946012169122696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07559830695390701,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05888964235782623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05035287141799927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04591505974531174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044816240668296814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029812291264533997,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024339772760868073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02380193956196308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021334176883101463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020627474412322044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016428286209702492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016223173588514328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014982105232775211,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011625177226960659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05035287141799927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05035287141799927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10928948223590851,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10006524622440338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09303509443998337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08444321900606155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.050899043679237366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04556712508201599,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.0658453032374382,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05945511534810066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05192425101995468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.044911228120326996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0437929630279541,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03360670059919357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0284200981259346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024586215615272522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023609735071659088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016865352168679237,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012933803722262383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012287275865674019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011757158674299717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011147325858473778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009004323743283749,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008747897110879421,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007616496179252863,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005942426156252623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05192425101995468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05192425101995468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09122080355882645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0827295333147049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07359453290700912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06687337160110474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04202055186033249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03561447188258171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.058935899287462234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053484562784433365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04306158050894737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03705049678683281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03659228980541229,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02994987927377224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0255876611918211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02033252641558647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01890396699309349,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014943400397896767,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010692759416997433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009786752983927727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009723183698952198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008815012872219086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007780096028000116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00760724488645792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005941588431596756,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004923895932734013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053484562784433365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053484562784433365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.21679089963436127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19406531751155853,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18209117650985718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16445887088775635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1000591441988945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08938778936862946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12616044282913208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11263936012983322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.1027204692363739,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08627913892269135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08350113779306412,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06491731852293015,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053830377757549286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04797002300620079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04648280143737793,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03252711519598961,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.024591311812400818,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02347838319838047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.021786458790302277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.020803524181246758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01671263948082924,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015561647713184357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014025588519871235,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00958433747291565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053830377757549286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053830377757549286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2303474247455597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.200810506939888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19055958092212677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16070528328418732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10687829554080963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09497074782848358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12309370189905167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11327061057090759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10911892354488373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08520635962486267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07752732932567596,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06319588422775269,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054441068321466446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05147664248943329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05077638849616051,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03168822452425957,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.026984339579939842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026227418333292007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02272414043545723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.022253308445215225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01691633090376854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.017261743545532227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01593218930065632,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012141730636358261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05147664248943329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05147664248943329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14371661841869354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13515472412109375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13193932175636292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12005634605884552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06801272928714752,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0648375153541565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07643576711416245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07048289477825165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06867460906505585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.060972027480602264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.058374855667352676,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03906713053584099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03381593897938728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03271247446537018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.032450467348098755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.01957164704799652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017039747908711433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.016797924414277077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015634218230843544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015469217672944069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010480672121047974,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010633102618157864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01012298185378313,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007406961638480425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03906713053584099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03906713053584099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23519429564476013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22171618044376373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21705223619937897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19742456078529358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1113191619515419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10647161304950714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12421892583370209,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11450593173503876,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11228305846452713,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09994170069694519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0955977588891983,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06342466175556183,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05481244996190071,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053373727947473526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05303028225898743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03174586966633797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0274603720754385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02712080255150795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025146547704935074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024931227788329124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01685454323887825,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016577906906604767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01638275384902954,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010981686413288116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053373727947473526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053373727947473526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.21.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2052726000547409,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18048812448978424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.16935867071151733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15033024549484253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0936426892876625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08314555883407593,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11404795944690704,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10393796861171722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09687253832817078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07892251014709473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07514623552560806,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.057932473719120026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049950361251831055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04526190087199211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044099144637584686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029377024620771408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02403027005493641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023432869464159012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021204207092523575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020439133048057556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016207829117774963,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016114214435219765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014700859785079956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011512459255754948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049950361251831055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049950361251831055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10957325994968414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09994599968194962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09271310269832611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08412954956293106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.050933849066495895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.045396242290735245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06604039669036865,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05968862771987915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.052117422223091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.044849272817373276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04372585937380791,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03375091031193733,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02858457900583744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024624032899737358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023603098466992378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01696127839386463,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012966522015631199,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012304143980145454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01177222654223442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011138373054564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009057406336069107,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008805310353636742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007635244634002447,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005993344821035862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.052117422223091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.052117422223091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09877453744411469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08956434577703476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07973670959472656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0725289136171341,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.045579712837934494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03869170323014259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06360765546560287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05781104788184166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04672587662935257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.040120042860507965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03957948833703995,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.032331496477127075,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027652788907289505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02208142727613449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020574579015374184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016180474311113358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011671765707433224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010717642493546009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010614946484565735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009668298065662384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008505980484187603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00833604484796524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006594098638743162,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0055103180930018425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04672587662935257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04672587662935257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2146129459142685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19117869436740875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1782289296388626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16099302470684052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09880749136209488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08744488656520844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12621107697486877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11253862082958221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10159089416265488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08476818352937698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08218155801296234,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06464135646820068,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053708694875240326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04738851636648178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04576430097222328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.032327935099601746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.024311577901244164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.023111021146178246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02147120237350464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02039506658911705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01654907315969467,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01550542563199997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013749765232205391,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009554313495755196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053708694875240326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053708694875240326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21571846306324005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18611060082912445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17531917989253998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14945052564144135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.1008707731962204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09003767371177673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11816170811653137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10780710726976395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.1029682606458664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07973866164684296,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0729612410068512,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06060759350657463,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05197511985898018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04882344603538513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04804620519280434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.030374085530638695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02580353058874607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025104764848947525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.021735278889536858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02121789939701557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016415836289525032,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016898665577173233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015385137870907784,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0122162364423275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05197511985898018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05197511985898018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15724799036979675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14801055192947388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14456278085708618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13154108822345734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07440970838069916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07095904648303986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08344663679599762,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07703141123056412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07512904703617096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06670434027910233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06390923261642456,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0426163449883461,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.036944154649972916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.035757336765527725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035480011254549026,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021316412836313248,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01857135072350502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018309658393263817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017029915004968643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01685103215277195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011337440460920334,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01150318793952465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010952526703476906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007916001603007317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0426163449883461,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0426163449883461,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23904350399971008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2253643125295639,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2206437885761261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20078004896640778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11306136846542358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10818864405155182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12597690522670746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11627401411533356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11410269141197205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1015218049287796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09717012196779251,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06427565962076187,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05563081428408623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05418301373720169,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05384523794054985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0321015864610672,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0277982447296381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02745252102613449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025441676378250122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02522149682044983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016887299716472626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01666787452995777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016404885798692703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010895858518779278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05418301373720169,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05384523794054985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.22.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21165591478347778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18738839030265808,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17656877636909485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1568971425294876,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0969102680683136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0867493748664856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11748203635215759,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1068744957447052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09991614520549774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08211133629083633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07816799730062485,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05971162021160126,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05125107616186142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0467614084482193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04563642665743828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.0302292350679636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024670062586665154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02412590943276882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021808689460158348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021095577627420425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016624247655272484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01631348580121994,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015167895704507828,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011537257581949234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05125107616186142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05125107616186142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11144105345010757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10212317854166031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09526549279689789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0864814817905426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05190658941864967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04660457372665405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06652797758579254,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06018252298235893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0529869981110096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04579433798789978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.044573768973350525,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.033835429698228836,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028793832287192345,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025053590536117554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02409978397190571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016952916979789734,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013133362866938114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012500988319516182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011926903389394283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011330543085932732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00897444412112236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008796442300081253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007630597334355116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005924706347286701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0529869981110096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0529869981110096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09834861755371094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08901286870241165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08058246970176697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07312973588705063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04527607560157776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03916103392839432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06106571853160858,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0556943379342556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04643521085381508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.039720963686704636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03891979157924652,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03095146454870701,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02656635455787182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02190260961651802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020674949511885643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.015520544722676277,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011497283354401588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010694186203181744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010401714593172073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.00962082203477621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008115554228425026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00796891562640667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006500585936009884,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005228078458458185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04643521085381508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04643521085381508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22936303913593292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20820878446102142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19763407111167908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17858092486858368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.106652170419693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0970284715294838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1297520250082016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11741314083337784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10894081741571426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09268859773874283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0893661305308342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06626816838979721,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.056016065180301666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05111228674650192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04990869015455246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033119283616542816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026141654700040817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025206558406352997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023295940831303596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02249949984252453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016994085162878036,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016197707504034042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014908474870026112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01002773828804493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05111228674650192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05111228674650192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2108493149280548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18377168476581573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17388375103473663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14398780465126038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0971338301897049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08699032664299011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11393299698829651,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10438046604394913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09974299371242523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07697572559118271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07006049156188965,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05831246078014374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05016935244202614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04685966670513153,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04606541618704796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.029204130172729492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.024611441418528557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.023848094046115875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02069205790758133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02015066333115101,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015600213780999184,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015930181369185448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.014527887105941772,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011241447180509567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05016935244202614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05016935244202614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16388827562332153,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1541646271944046,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15059228241443634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13702112436294556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07755082100629807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0739508718252182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08709289133548737,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08020004630088806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07829032838344574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06949421763420105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0665753185749054,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04442601650953293,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.038472916930913925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037278834730386734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03699669614434242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02231348305940628,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019385285675525665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01911500096321106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.0177652295678854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017593974247574806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012034471146762371,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012025880627334118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01165598165243864,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008325144648551941,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04442601650953293,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04442601650953293,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24376733601093292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22961558401584625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2248132824897766,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20449243485927582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11534594744443893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11028263717889786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12855149805545807,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11855839192867279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11639999598264694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10343199223279953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09889744967222214,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06556588411331177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05673935264348984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05528217926621437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05493824928998947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03280350938439369,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028353199362754822,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027995241805911064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025915885344147682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025698309764266014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017327969893813133,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016974076628684998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01683942787349224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011072498746216297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03280350938439369,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03280350938439369,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.23.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21763044595718384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1923941969871521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18110594153404236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16124969720840454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09962042421102524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08897606283426285,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.120120570063591,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11001227796077728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10273993015289307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.0843755230307579,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08043208718299866,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.061254099011421204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05274949595332146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0480387881398201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04686591029167175,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030863473191857338,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025283673778176308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02469497174024582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022322557866573334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021569600328803062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01666448265314102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01665089838206768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015107650309801102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011665080673992634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05274949595332146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05274949595332146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11982377618551254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1102212518453598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10303369909524918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09354404360055923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05602928623557091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05050589516758919,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07108020782470703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06468871235847473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05703223869204521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.049530621618032455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.048145826905965805,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.036255158483982086,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030983267351984978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02705051563680172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.026059577241539955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018155114725232124,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014207431115210056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013546726666390896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012927585281431675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012312479317188263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009571581147611141,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009526669979095459,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008186751045286655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006474317982792854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.049530621618032455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.049530621618032455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.105857253074646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.09681237488985062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08756309002637863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07963361591100693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.049108684062957764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0425257682800293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0668167695403099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.060869138687849045,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.050196968019008636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04343812167644501,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04267967864871025,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03392903879284859,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.029076386243104935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0237272996455431,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022276148200035095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016949808225035667,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012385053560137749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011477666907012463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011267125606536865,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010363537818193436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008854063227772713,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008587241172790527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007035992108285427,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005521143786609173,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.050196968019008636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.050196968019008636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23770567774772644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21765241026878357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.2083035111427307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1884317398071289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11104048788547516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.1023450717329979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13261345028877258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12016896903514862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11312233656644821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09735787659883499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09344839304685593,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06752979010343552,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05735821649432182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053194575011730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05217171087861061,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03375744819641113,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027144575491547585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02631070837378502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02430320531129837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023617621511220932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017297660931944847,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016562573611736298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015533704310655594,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01016708742827177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053194575011730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053194575011730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1822643131017685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1650846302509308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15686260163784027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1310717761516571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0854109600186348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0780898779630661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10173016786575317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09326981008052826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08658351749181747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07047121971845627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0642717108130455,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05195440351963043,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04468100890517235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04109637439250946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04020283371210098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.025976350530982018,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021295061334967613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.020598160102963448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018290936946868896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.017697649076581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013676178641617298,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.013509802520275116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012464023195207119,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00896008126437664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05195440351963043,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05195440351963043,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1644883006811142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1546863466501236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15115080773830414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13738568127155304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07785758376121521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07422785460948944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08722874522209167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08050209283828735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07861840724945068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06969896703958511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06669887900352478,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044481806457042694,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03859532251954079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03740832582116127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.037124037742614746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02226800099015236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01938902959227562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01912275142967701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017748164013028145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017569800838828087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011835074983537197,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01194112841039896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01145770400762558,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008156314492225647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044481806457042694,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044481806457042694,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24699078500270844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23270951211452484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2278493493795395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20718933641910553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11691351979970932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.1118030995130539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13060861825942993,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12017872929573059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11802347749471664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10484223067760468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.1002972275018692,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06657490879297256,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.057536207139492035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05607863888144493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.055738404393196106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033394187688827515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028808271512389183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02845500037074089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026330342516303062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026110786944627762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01778838224709034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01731196418404579,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01731100305914879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011376766487956047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033394187688827515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033394187688827515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.24.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2215447723865509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19618546962738037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1850007325410843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16475047171115875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1013823002576828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09081171452999115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12227802723646164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11167708784341812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10460495948791504,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08610530942678452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0819515660405159,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06198311969637871,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05348736792802811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04883664473891258,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04767434298992157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.031283020973205566,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025610540062189102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02503371611237526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022622372955083847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02187088504433632,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016980202868580818,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016727423295378685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01545341033488512,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0115788159891963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05348736792802811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05348736792802811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12556305527687073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1159956231713295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10893036425113678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09871136397123337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0588366873562336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05336252599954605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07406643033027649,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06730064749717712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.059843894094228745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.052128471434116364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05057583749294281,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03771689906716347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.03223186358809471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02836601994931698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.027398614212870598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018879787996411324,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014800465665757656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01413826085627079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013466855511069298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012852605432271957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00989446323364973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00976674072444439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00851232185959816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006496401038020849,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.052128471434116364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.052128471434116364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10687704384326935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.09802205115556717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08918632566928864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.08103649318218231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.049645498394966125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.043320197612047195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0663713738322258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.060791224241256714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05062929913401604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04395134374499321,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04291936010122299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03364581987261772,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02906634286046028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02394874580204487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022595783695578575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016815830022096634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012461531907320023,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011590085923671722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01133390050381422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010473296977579594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008740531280636787,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00853308942168951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007042492739856243,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00542002497240901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05062929913401604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05062929913401604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.239474356174469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21975542604923248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.2106943130493164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1905088871717453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11187077313661575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10350967943668365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1340796798467636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12072005122900009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11383438855409622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0981183797121048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09425124526023865,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06833413988351822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0576237253844738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053602997213602066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.052614402025938034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.034064799547195435,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02730974368751049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02648700773715973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.024464676156640053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023826181888580322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01750899851322174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016577132046222687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01571711339056492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010110314004123211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053602997213602066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053602997213602066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19715285301208496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17291495203971863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1650565266609192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.136726513504982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09188529849052429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08323013037443161,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10596141219139099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09645688533782959,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09369111061096191,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07316645234823227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06495869904756546,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05389346927404404,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04622245207428932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04414921998977661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04363996908068657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027014337480068207,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.022814135998487473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02226768620312214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01903679035604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018694404512643814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014341474510729313,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014087200164794922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013615787029266357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.009435413405299187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05389346927404404,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05389346927404404,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16994985938072205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15970930457115173,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15601615607738495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14184722304344177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0805244967341423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07675040513277054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0902675986289978,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08324528485536575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08130747824907303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07200834900140762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06889088451862335,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046129852533340454,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03994375467300415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03870568424463272,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03841806575655937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.023116545751690865,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020098570734262466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019830169156193733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018395040184259415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01821562834084034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01236339844763279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012440857477486134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011966140940785408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008563359268009663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046129852533340454,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046129852533340454,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.25007277727127075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23540708422660828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2304924726486206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.209419846534729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11847733706235886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11326760798692703,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13172279298305511,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12175863236188889,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11957496404647827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10608039051294327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.10131092369556427,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06711410731077194,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05823390185832977,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05675370618700981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.056398868560791016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03351031243801117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.029027877375483513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02866833098232746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026475690305233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026250038295984268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017450524494051933,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017264489084482193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016953660175204277,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011098220013082027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03351031243801117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03351031243801117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.25.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22341962158679962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19779229164123535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18608558177947998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16591614484786987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10212691873311996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09126339107751846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12343217432498932,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11322055757045746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10547378659248352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08701349794864655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08286767452955246,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06288120150566101,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.054267920553684235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.049180008471012115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04791262000799179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03156855329871178,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02575099654495716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025119254365563393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02278914488852024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021966073662042618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0168871209025383,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01683790795505047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015216715633869171,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0115489661693573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.054267920553684235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.054267920553684235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12053371965885162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.11076760292053223,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10318152606487274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.093513622879982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.056388288736343384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05059025064110756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07188374549150467,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06559693813323975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05744076520204544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04981626942753792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.048390377312898636,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03664573282003403,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.031455881893634796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.027232738211750984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.026162054389715195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018347304314374924,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014282265678048134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013571036979556084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012987343594431877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012313530780375004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00958429928869009,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009602971374988556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008095034398138523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00645834906026721,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04981626942753792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04981626942753792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10353386402130127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0944262221455574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08451580256223679,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07670380175113678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04793551564216614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.04102899506688118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06642840802669525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.06052868440747261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04905562102794647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.042328137904405594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.041505519300699234,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0335739329457283,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02885543182492256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02319573424756527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.021646760404109955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016795940697193146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012149459682404995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011178635992109776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011039599776268005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010068141855299473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008812550455331802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008530905470252037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006894449237734079,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005481294821947813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04905562102794647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04905562102794647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23001907765865326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20809012651443481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1968235969543457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17770859599113464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1067451760172844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09663660824298859,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13227549195289612,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11854508519172668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10938778519630432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09268905967473984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08947689831256866,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06777966022491455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05658711493015289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123372748494148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049881577491760254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033908549696207047,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026182422414422035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02514859102666378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023289721459150314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02240009233355522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017397766932845116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016305483877658844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014977026730775833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009975658729672432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123372748494148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123372748494148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18919837474822998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17131733894348145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16575421392917633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14084136486053467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08916810154914856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08256521075963974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10014063864946365,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09214363992214203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09038402140140533,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07386420667171478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06727684289216995,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.051175300031900406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.044151365756988525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04279990494251251,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04248662292957306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.025581825524568558,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02205626666545868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.021664096042513847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018968136981129646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018748996779322624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013436826877295971,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.013422160409390926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012991671450436115,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.008961319923400879,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.051175300031900406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.051175300031900406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17453797161579132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.16400179266929626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.16019724309444427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14550961554050446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08278516680002213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07890249043703079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09273967146873474,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08559832721948624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0836155116558075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0739915668964386,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07065751403570175,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04734564572572708,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.041043464094400406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03977876156568527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.039477959275245667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.023681458085775375,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020611200481653214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020329318940639496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018839513882994652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018649715930223465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012530026957392693,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012689477764070034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01211528293788433,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008654211647808552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04734564572572708,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04734564572572708,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2512197196483612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23645330965518951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.23150409758090973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.21027915179729462,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11918137967586517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11390683054924011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13286268711090088,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12249982357025146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.12028618156909943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10664454847574234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.10184229910373688,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06776771694421768,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0586634986102581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.057152822613716125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0567997507750988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033883705735206604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.029367459937930107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.029003577306866646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026795823127031326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026569750159978867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01788564771413803,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01764919050037861,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01738482527434826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011606301181018353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033883705735206604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033883705735206604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.26.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22692427039146423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20162063837051392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19002698361873627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1692453920841217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10395738482475281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09318237006664276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12549404799938202,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11499937623739243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10729505121707916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08877211809158325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08449645340442657,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06395233422517776,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0551777146756649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05010344088077545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04885312542319298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03233486786484718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026354603469371796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025724049657583237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023392533883452415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022584237158298492,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017627809196710587,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017340756952762604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015992915257811546,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012071800418198109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05010344088077545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05010344088077545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12452154606580734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.11453695595264435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10809440165758133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09805474430322647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05838008224964142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.053114622831344604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07246813923120499,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06587137281894684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05941365286707878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05149584263563156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0498381033539772,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03697812929749489,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.031559258699417114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.028150558471679688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.027310607954859734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01852261647582054,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014705331064760685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01410877425223589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013334398157894611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012790704146027565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009739626199007034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009637013077735901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008491975255310535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0064606983214616776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05149584263563156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05149584263563156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10733161121606827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0989459753036499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.09235842525959015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0837026834487915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.05014695227146149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.04509536176919937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06361188739538193,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05822722986340523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05106978118419647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04435816779732704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04289370775222778,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03225512057542801,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02784596011042595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.024184204638004303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.023253321647644043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016180627048015594,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012574711814522743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011957300826907158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011417492292821407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010832133702933788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008471757173538208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008316163904964924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007201797794550657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005438519176095724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05106978118419647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05106978118419647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22618617117404938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20783376693725586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19920216500759125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.18024316430091858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10609455406665802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09800407290458679,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.125311478972435,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11450386047363281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10792074352502823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09308404475450516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08924183249473572,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06385636329650879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05472434312105179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05085771530866623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04990743100643158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.031863462179899216,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02589772827923298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02509617619216442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023191772401332855,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022561104968190193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01634899154305458,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015685463324189186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014880356378853321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009533638134598732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05085771530866623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05085771530866623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.20709311962127686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18371646106243134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1756037026643753,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14739514887332916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09714093059301376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08835341036319733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11171499639749527,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10243193060159683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09894679486751556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07798339426517487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07212778180837631,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.057404011487960815,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04943528771400452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04704694449901581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.046497926115989685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02893695794045925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02502412721514702,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02444585971534252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.021348239853978157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020963555201888084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015841927379369736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016442058607935905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01509474590420723,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012123816646635532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04943528771400452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04943528771400452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1696636974811554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15925085544586182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1554262638092041,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14101669192314148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0805472731590271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07662193477153778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09046194702386856,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0834740698337555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08137940615415573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07186915725469589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06864246726036072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04620400816202164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04008869826793671,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.038752127438783646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.038441333919763565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02313356287777424,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02013331465423107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019838061183691025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01839279942214489,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018199166283011436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012288039550185204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012490564025938511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011849011294543743,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00861911941319704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04620400816202164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04620400816202164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.243917778134346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22943468391895294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22448599338531494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20380203425884247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1157817617058754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11056102067232132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12916427850723267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11916787177324295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11689942330121994,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1034947857260704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09887633472681046,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06598954647779465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05710281804203987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.055572204291820526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05520958453416824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0330083966255188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028569214046001434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0281955786049366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02604973316192627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025813542306423187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017437513917684555,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017210720106959343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016923565417528152,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011341173201799393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0330083966255188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0330083966255188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.27.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22700971364974976,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20089219510555267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18851624429225922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16792292892932892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10385089367628098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09249718487262726,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1275073140859604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11588882654905319,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10735159367322922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08841566741466522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0842674970626831,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06474127620458603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.055522285401821136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05008291080594063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04873017966747284,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.032673366367816925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026347137987613678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025668693706393242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023331694304943085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022461935877799988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017774414271116257,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017407815903425217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015970418229699135,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012096922844648361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05008291080594063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05008291080594063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11572492867708206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10626456886529922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09801332652568817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08891233056783676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.054125506430864334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04799175262451172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07131495326757431,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0644075870513916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05519085004925728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04779224097728729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04679208621382713,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03645377233624458,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030863337218761444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02618774026632309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024978477507829666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01823790743947029,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013802701607346535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013025574386119843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012560619041323662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011811941862106323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009593253023922443,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009462372399866581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007903127931058407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0064032673835754395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.054125506430864334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.054125506430864334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09731120616197586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08851464837789536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07734818011522293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07040656358003616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04492749646306038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.037374380975961685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06498654931783676,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.059081148356199265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04605111479759216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.039760228246450424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03935680538415909,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03285251930356026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02815963886678219,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021789396181702614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020010769367218018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016488082706928253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011497395113110542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.01039078924804926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010493731126189232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009369997307658195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008600966073572636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008353759534657001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006392995826900005,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005371517036110163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04605111479759216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04605111479759216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23069420456886292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2078019231557846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19543907046318054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17634005844593048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10710982978343964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09611407667398453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1343490481376648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1203572005033493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10976240783929825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0927569717168808,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0896335318684578,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06881999224424362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05757840350270271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05147475376725197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049937356263399124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.034555140882730484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026490207761526108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02535529062151909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023588579148054123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022594979032874107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017789918929338455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01691047102212906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015112156048417091,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010707507841289043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05147475376725197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05147475376725197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21962280571460724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18841341137886047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1776753067970276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15232819318771362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10183048993349075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09018845111131668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.1187899112701416,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10850027948617935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10426682978868484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07932937145233154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07580279558897018,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06088311970233917,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.052138980478048325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04907821863889694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04832329601049423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03049134463071823,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.025779221206903458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.024988699704408646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02140943519771099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020911216735839844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016389766708016396,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01662471890449524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015413561835885048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011755919083952904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.052138980478048325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.052138980478048325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15797848999500275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1480463296175003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14414793252944946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1307307481765747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07497101277112961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07113786041736603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08464109897613525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0781075581908226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07579217851161957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0668020099401474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06375230103731155,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04328674077987671,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03749806433916092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03608153015375137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035743407905101776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02166181430220604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018741797655820847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01844456046819687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01710067130625248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01689102128148079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011502068489789963,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011650791391730309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011037035845220089,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008009687066078186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04328674077987671,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04328674077987671,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22683678567409515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2130274772644043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2081146389245987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18886259198188782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10766664892435074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10252943634986877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1204918920993805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11119615286588669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10873465240001678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09609802812337875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09165717661380768,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061596956104040146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053289253264665604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0517064668238163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0513235367834568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03083229809999466,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0266769677400589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02631176821887493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024330344051122665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02409081533551216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01633727364242077,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01625257357954979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015798402950167656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010893873870372772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053289253264665604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053289253264665604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.28.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20614174008369446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1810261756181717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.168910950422287,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.14987972378730774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09427877515554428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08322176337242126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1160864531993866,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10599948465824127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09758887439966202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07945822179317474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07577288150787354,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05936186760663986,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.051070112735033035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.045723121613264084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04437762871384621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030095098540186882,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024472499266266823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023798203095793724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021657794713974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020795727148652077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01661045290529728,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01673768274486065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014861950650811195,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012166278436779976,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.051070112735033035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.051070112735033035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11367612332105637,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10432964563369751,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09678502380847931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08786752074956894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053042247891426086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04731567203998566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07015542685985565,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06234882026910782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05415944382548332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04688425362110138,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04583379998803139,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.035805508494377136,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.029868802055716515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025623805820941925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024538956582546234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01795237883925438,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013443049974739552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012744519859552383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01222216710448265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011546935886144638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009424896910786629,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009090684354305267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007751536555588245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006098092067986727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05415944382548332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05415944382548332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0973631888628006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08995675295591354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08168809115886688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07412026077508926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04534174129366875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.039626553654670715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06180315464735031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05597155541181564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04617545008659363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.040325410664081573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.039512503892183304,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03140662983059883,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02674219384789467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021911388263106346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020605646073818207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.01570458523929119,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01145564578473568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010617943480610847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010464434511959553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009630531072616577,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00819715578109026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007936128415167332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006453374866396189,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005081410985440016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04617545008659363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04617545008659363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.24874486029148102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2292284071445465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.22003290057182312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1993403136730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11686625331640244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10830976068973541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13931411504745483,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1263684183359146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11889489740133286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.10296541452407837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09905707836151123,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.07132161408662796,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.06042494252324104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05609932541847229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.055037982761859894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.035800572484731674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.028698571026325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.027871983125805855,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.025843419134616852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.025148339569568634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01840903051197529,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01761748641729355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.016524985432624817,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.011030805297195911,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.035800572484731674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.035800572484731674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18839126825332642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.16278888285160065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15342433750629425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.13441592454910278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08790618926286697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07792135328054428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10402284562587738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09350671619176865,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08984015882015228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06998612731695175,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06666959822177887,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05339683219790459,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04496820643544197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.042426321655511856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04180406033992767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026778001338243484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.022253278642892838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.021577540785074234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018739059567451477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.01832115463912487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014231531880795956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014290078543126583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013338725082576275,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01006705779582262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05339683219790459,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05339683219790459,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15341390669345856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14361542463302612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13970118761062622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1266121119260788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07276876270771027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06895560771226883,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08237628638744354,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07598202675580978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07361187040805817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06476320326328278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.061901893466711044,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04211081936955452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03650331497192383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03506097197532654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03471751883625984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021085144951939583,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018305864185094833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018002627417445183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.016710253432393074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.016491355374455452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01124162133783102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011525582522153854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01075770240277052,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008063158951699734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04211081936955452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04211081936955452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.20209452509880066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1895570456981659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.184932142496109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1678933948278427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09669799357652664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09197380393743515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.10844431817531586,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10006734728813171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09762997180223465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08626969903707504,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08249043673276901,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.056075092405080795,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04892962425947189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.047404706478118896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.047041989862918854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.028243333101272583,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0261384230107069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025827614590525627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02423454262316227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024021128192543983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.015970636159181595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.018166454508900642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01549572590738535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.014622834511101246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04892962425947189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04892962425947189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.29.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20571546256542206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1815090775489807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17045757174491882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1500616818666458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09432607889175415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08405863493680954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1150156781077385,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10463142395019531,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09726862609386444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07922500371932983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07495211809873581,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05831105634570122,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05018393322825432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.045558180660009384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04440474137663841,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029313212260603905,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02412106655538082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02354668453335762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02121189422905445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02046799287199974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01579117961227894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016077060252428055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014262380078434944,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.01144026592373848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05018393322825432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05018393322825432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1095728725194931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.100243479013443,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09128393232822418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08260037004947662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.051102083176374435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04463138431310654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06921995431184769,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06240910664200783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05219741538167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04499661177396774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04423733800649643,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0353984534740448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02993728592991829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02478559873998165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023427501320838928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017751535400748253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013142452575266361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01227540336549282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011954843997955322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011102452874183655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00936879962682724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009230590425431728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0075102150440216064,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006275800988078117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05219741538167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05219741538167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09547897428274155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08686858415603638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07536287605762482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0684126615524292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04400666058063507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03634760528802872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06484340876340866,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.058620940893888474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04517835006117821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03889043256640434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03865814954042435,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.032974984496831894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027953792363405228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021396392956376076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019549990072846413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016563745215535164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01135834027081728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.01020627561956644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010375534184277058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009200098924338818,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008627797476947308,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008383037522435188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006264181341975927,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005445764400064945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04517835006117821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04517835006117821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22474946081638336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20137616991996765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18781088292598724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16936330497264862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10393724590539932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09239887446165085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13313470780849457,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11901592463254929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10685907304286957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08965485543012619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08724135905504227,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0685306265950203,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05686498433351517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05003635585308075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04826974868774414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03444301709532738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025873634964227676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02458743378520012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.022994350641965866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021843496710062027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01781383715569973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016799015924334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01471610739827156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010716233402490616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05003635585308075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05003635585308075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.16379593312740326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.14769664406776428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1425313651561737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.12055926024913788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07679182291030884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07108142971992493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.08731945604085922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.07948681712150574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07775185257196426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06287302076816559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05801357328891754,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04490472376346588,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.03835875540971756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.03710135444998741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.036819204688072205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0225103497505188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01961367391049862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.019267059862613678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01692737452685833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.016717446967959404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.012195341289043427,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.012644082307815552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.011801408603787422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.009256135672330856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04490472376346588,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04490472376346588,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15331622958183289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14363162219524384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1398492157459259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12690052390098572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07327380031347275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0694596990942955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08264171332120895,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07623498886823654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07402262091636658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06524883955717087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06245270371437073,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04276455193758011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03727343678474426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03593238443136215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035613540560007095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02160659246146679,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019766774028539658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019496247172355652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01829102821648121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018106089904904366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012209240347146988,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013711347244679928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011786255054175854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01097308099269867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04276455193758011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04276455193758011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.161906898021698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15187588334083557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14817042648792267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1344291865825653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07737308740615845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07358581572771072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0868956670165062,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08015749603509903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07815074175596237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06899557262659073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06600962579250336,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04486967250704765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039062194526195526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037813249975442886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03752044960856438,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022616958245635033,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020649300888180733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02039312571287155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01909191906452179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01891881786286831,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012667782604694366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.014124766923487186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012267973273992538,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011174408718943596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04486967250704765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04486967250704765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.30.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.17595866322517395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.15610191226005554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1475788950920105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.12861225008964539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.08129768073558807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.07320835441350937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.099279023706913,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.0883098840713501,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.08329994976520538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.06792747229337692,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.06408126652240753,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05014503374695778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.042950693517923355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.03983841836452484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.03902662917971611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.025673769414424896,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.021892568096518517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02151186391711235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.019457319751381874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.01897076703608036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.014794301241636276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015404149889945984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.013819348998367786,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012092744931578636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05014503374695778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05014503374695778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1061895564198494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0968628078699112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0882173702120781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07984215021133423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04949919879436493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.043128401041030884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.066900834441185,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.060333918780088425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05060521513223648,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04350423440337181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04272889345884323,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03416188061237335,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028950683772563934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02398574724793434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022681990638375282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017103267833590508,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01269147265702486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011865245178341866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011522394604980946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010716697201132774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009003915823996067,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008881181478500366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007230670191347599,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006025675218552351,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05060521513223648,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05060521513223648,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09517841041088104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08642280846834183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07565269619226456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06861930340528488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04396772384643555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.036606211215257645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06458213925361633,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05770289897918701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.045150693506002426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.038752928376197815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.038485389202833176,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03274136036634445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027539081871509552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02137724496424198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019634725525975227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016404755413532257,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01133318804204464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010291951708495617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010328191332519054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009264237247407436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00861305184662342,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00828433409333229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006385622546076775,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005472538061439991,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.045150693506002426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.045150693506002426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2369954138994217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20976445078849792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19394990801811218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17515923082828522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10949000716209412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09581846743822098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.140801802277565,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12658214569091797,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11301255971193314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09350759536027908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09111417084932327,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.07231248170137405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.06061223894357681,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.052722275257110596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05070376768708229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03639683127403259,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027204643934965134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025729410350322723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023973781615495682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02264346554875374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.018712390214204788,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01772727072238922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015440641902387142,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01116263773292303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.052722275257110596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.052722275257110596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1016402393579483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09097951650619507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08753809332847595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07567872107028961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0461217425763607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04225469008088112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.054787587374448776,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04818283021450043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04686923325061798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0379980094730854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03609740734100342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02759561501443386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.024853995069861412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.023935209959745407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02371959760785103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.014634456485509872,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014866461046040058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.014681817963719368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013609597459435463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.013487079180777073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009156208485364914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.011941846460103989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00888765323907137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010768837295472622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04818283021450043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04818283021450043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14864933490753174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.139508455991745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13613034784793854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12340245395898819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07056599110364914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06713946163654327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07937955111265182,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07310084253549576,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07123634964227676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06288063526153564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06006962060928345,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.040562666952610016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.035055696964263916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0339076891541481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03362968564033508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02027907967567444,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01753162033855915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.017274271696805954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015979034826159477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01580420695245266,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010695081204175949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010762694291770458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010303606279194355,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007279439829289913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.040562666952610016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.040562666952610016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.10389047116041183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.09726250171661377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.09485975652933121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.08593438565731049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.04926689341664314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.04681713879108429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.05551265552639961,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.05106944218277931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.049770206212997437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.043839775025844574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.041905198246240616,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.028441954404115677,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.024597933515906334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.023779118433594704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02358659729361534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.014284290373325348,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.012519893236458302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.012341500259935856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.011453012935817242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01133302878588438,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.007762065157294273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.007997272536158562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.0074990964494645596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.005783628206700087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.05106944218277931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.05106944218277931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ },
+ {
+ "key": "model.layers.31.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.10926377028226852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.09788458794355392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.09272480010986328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.08060410618782043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.050845175981521606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.04602187126874924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.06256718933582306,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.05570872128009796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.05201271176338196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.042947325855493546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.040726907551288605,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.032257549464702606,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.027281779795885086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.025067172944545746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.024508720263838768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.016622059047222137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.014052143320441246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.01378138829022646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.012646831572055817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.01232102606445551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.009740615263581276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.010195295326411724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.009029214270412922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.008230828680098057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ],
+ "best_option_max": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.05201271176338196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ "best_option": {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.05201271176338196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ }
+ }
+ ],
+ "base_perplexity": 6.128923067853044,
+ "q_last_module_idx": 65
+}
\ No newline at end of file
diff --git a/measurement.json b/measurement.json
new file mode 100644
index 0000000000000000000000000000000000000000..8803c8019cd053fd58be4b77c537d8ff4624a28e
--- /dev/null
+++ b/measurement.json
@@ -0,0 +1,95878 @@
+{
+ "measurement": [
+ {
+ "key": "model.layers.0.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.011168720200657845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.009599272161722183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.004790821112692356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.005077867768704891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0050776307471096516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0020578650292009115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.01079169474542141,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.009518211707472801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.005259049125015736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.004644699394702911,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0048562814481556416,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0050615472719073296,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.004643067717552185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0027538840658962727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.002120051998645067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0026784869842231274,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.001900350907817483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.001615153276361525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0018467125482857227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.0015654281014576554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0017823567613959312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0018463897285982966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0014250442618504167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.001544640981592238,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.011349556967616081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.00980227068066597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.004773853346705437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.005033744964748621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.005033500958234072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.001928701065480709,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.01110624335706234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.009691532701253891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.005209838971495628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.004562461748719215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0047759003937244415,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.005032561253756285,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.004560141358524561,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.002655236516147852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.00194216996897012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0026120387483388186,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0016779541037976742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.001346124685369432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0016128767747431993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0012812796048820019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.001617392641492188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0016132771270349622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0012200467754155397,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0012539472663775086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.11578802764415741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07042404264211655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.043504536151885986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.050045717507600784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0499984472990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.024094626307487488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.07798509299755096,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.06458660215139389,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0545671209692955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03145388513803482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03755158931016922,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.042070720344781876,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.031096184626221657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02426334097981453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022416630759835243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02171494998037815,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01284871157258749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011072240769863129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009356767870485783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007840869016945362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.011130761355161667,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.009300582110881805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007214382756501436,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.006086327601224184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11313514411449432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0732056200504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05039582401514053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04979194700717926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.048297327011823654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.027966083958745003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07069999724626541,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06314126402139664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.053515732288360596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.031263019889593124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03367261961102486,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03614491969347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030515408143401146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024038290604948997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022272668778896332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0183316171169281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013834427110850811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012738020159304142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011082837358117104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009938649833202362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01025769766420126,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.010982340201735497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00825857650488615,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0086810402572155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.11327356100082397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1058005541563034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.10340237617492676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.09472587704658508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.051241014152765274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.04909159988164902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.05705980956554413,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.052599381655454636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.051703788340091705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.046595051884651184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.04498507082462311,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.029025588184595108,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.02526126801967621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.024675775319337845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02453121356666088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.01461299229413271,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.013298182748258114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.013174419291317463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.012456698343157768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.012386062182486057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.008203603327274323,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.008870589546859264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.008029122836887836,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.006887989118695259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.13404832780361176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.12670819461345673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12447718530893326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.114055335521698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06098993122577667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.05894749239087105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.06735212355852127,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.06225910410284996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06142910569906235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.055878788232803345,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05380960926413536,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03398489952087402,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.029508553445339203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.02898118458688259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02886049821972847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.016938773915171623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01486075110733509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.014734474010765553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01386601198464632,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.013790595345199108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.008871518075466156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.008892863988876343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.008702414110302925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.005870731081813574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.0.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.07670606672763824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.06584762781858444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.06038632243871689,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.05355939269065857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.03438195586204529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.029740920290350914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.04498032480478287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.03952185437083244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.03574554622173309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.028567753732204437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.02737824246287346,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.022440247237682343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.019085902720689774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.01691751927137375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.016391701996326447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.011539027094841003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.00958372000604868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.009348219260573387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.008578285574913025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.008279938250780106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.006880487315356731,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.00712636299431324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.00626949081197381,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0057810521684587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.019663630053400993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.013406879268586636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.007660592906177044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.008349855430424213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.00809794757515192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0036465248558670282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.013933522626757622,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.012625516392290592,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0092203663662076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.006169123109430075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.006603498011827469,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.007054396439343691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.006034728605300188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0040495740249753,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.003436762373894453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0035601980052888393,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0023379670456051826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.0019536123145371675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0020181608851999044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.001642401795834303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0019508769037202,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0019979930948466063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0013339562574401498,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0014734480064362288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.017518868669867516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.01219908520579338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.006678944453597069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.007290045265108347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0070999860763549805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0029782995115965605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.013018052093684673,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.011648166924715042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.008138769306242466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.005624793935567141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.006064086686819792,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0065219649113714695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.005529758520424366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0035434754099696875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.0028902385383844376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.003280432429164648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0020198565907776356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0016353140817955136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0017707019578665495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0013762976741418242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0017502045957371593,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0017561280401423573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.001087298383936286,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0012416673125699162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.13320785760879517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08857599645853043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06414425373077393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.062488194555044174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.057051729410886765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.034113503992557526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.08158433437347412,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.07376103848218918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0631725862622261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03857894614338875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.039744194597005844,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04156964644789696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.03522882238030434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.027669740840792656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.02555888704955578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.020771265029907227,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.014558087103068829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.012766730040311813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011037899181246758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009418193250894547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.010778137482702732,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.010443516075611115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.008073403500020504,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.006722630932927132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.15270331501960754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.12318339943885803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.11086562275886536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09390351921319962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07004237920045853,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05806812644004822,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.08862987905740738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.07844515144824982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07241526991128922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05164502561092377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.049597982317209244,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04550931602716446,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0382709726691246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.034586384892463684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.033655546605587006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02309839054942131,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01936962828040123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01856229268014431,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01617896556854248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.015599473379552364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013060690835118294,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014155326411128044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.011925097554922104,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011372439563274384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1567811667919159,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1485101878643036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14568355679512024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13279618322849274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0726742148399353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07009439915418625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0808241069316864,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0739935040473938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07315933704376221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06607392430305481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06361325085163116,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04134838655591011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03595521301031113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03534533828496933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03519531339406967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021018361672759056,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019455352798104286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019302669912576675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01826331578195095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01818668842315674,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012167708948254585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013381781987845898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012010962702333927,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010793034918606281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.19255827367305756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.18272824585437775,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.17977085709571838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.16407370567321777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08971576392650604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.08672579377889633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.098969966173172,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.09122589230537415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09027554839849472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08172681927680969,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0785735547542572,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.050545584410429,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.043932583183050156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04328686743974686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04312015324831009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.025338059291243553,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02311822585761547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02295522764325142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02161826193332672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02152479812502861,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.013921397738158703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015096379444003105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.013726180419325829,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011499397456645966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.1.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.019199078902602196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.01878763735294342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.006921666674315929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.00646975776180625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.005650496110320091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0034061234910041094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.018811136484146118,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.018254732713103294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.005731276702135801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.005354026332497597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0053440118208527565,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.005248498171567917,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.005061940755695105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0026619865093380213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.0026766916271299124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.002273819176480174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.0023983637802302837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.001029878854751587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.0023623292800039053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.000951776746660471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0023449964355677366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.0023288021329790354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.0007354797562584281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0007976787746883929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.05501016974449158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.043334461748600006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.03665383905172348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.03312649205327034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.024345438927412033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.018490733578801155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.03337150067090988,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03038576804101467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.025987472385168076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.018868857994675636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.018617024645209312,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.01694329082965851,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.014517929404973984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.011795316822826862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.011069130152463913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.008465803228318691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.00624634325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.005706323776394129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.005192603450268507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.004691623616963625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00441138306632638,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.004422945436090231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0034645930863916874,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0029624789021909237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.05672017112374306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04279809445142746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.03405015915632248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.03148718550801277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.024548310786485672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.017191600054502487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03530892729759216,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.032226257026195526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.026598580181598663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.01862349361181259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.018713288009166718,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.017903584986925125,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.015325364656746387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.011925559490919113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.010991599410772324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.008935020305216312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0063719660975039005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.005681759677827358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.0052579911425709724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.004604979418218136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.004677099175751209,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004683797247707844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0035144537687301636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00315110944211483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.16881753504276276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1410287767648697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.12958964705467224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.11464769393205643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.07729382812976837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.06546717882156372,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09526020288467407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08619674295186996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08046350628137589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.061131030321121216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.058465443551540375,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04843978211283684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.041259944438934326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.03719085827469826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03618834540247917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02420763485133648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01911242865025997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.018178023397922516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.015842380002141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.015133141539990902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012576712295413017,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012094818986952305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011064727790653706,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0076539963483810425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.156541645526886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.13858985900878906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.12967531383037567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.11246905475854874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07320044189691544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.06499434262514114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09032202512025833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.08170590549707413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07497423142194748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06085894629359245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05692308768630028,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04705378785729408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0400383360683918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.036137260496616364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.035172898322343826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024017736315727234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.020189229398965836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.019475512206554413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0179683156311512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.017397213727235794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013736758381128311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014681994915008545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01252848468720913,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01172161940485239,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.19975252449512482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.18823857605457306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1846926212310791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.16731983423233032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09383014589548111,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09011030942201614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.10378500074148178,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.09576664119958878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09457574784755707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08421718329191208,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08041726052761078,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05310473218560219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04596571624279022,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04511168226599693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04491891711950302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02657609060406685,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.023638196289539337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.023406749591231346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.021765321493148804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.021637681871652603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.014316429384052753,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.014850785955786705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.014048966579139233,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010670391842722893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23413872718811035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22090375423431396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21686704456806183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19659093022346497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10993479937314987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10565081238746643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12136806547641754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.1120128333568573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11077161878347397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09870945662260056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0940827876329422,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06181109696626663,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05351909250020981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05259128287434578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.052355699241161346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03086814470589161,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02699517272412777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02673065848648548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024733295664191246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024595467373728752,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016258612275123596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016147766262292862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015950508415699005,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010682974942028522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.2.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.1790362149477005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.15777480602264404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.14935097098350525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.13222238421440125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.08091238141059875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.07288438826799393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.0961962640285492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.08775460720062256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.083379827439785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.06838435679674149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.06484240293502808,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.04880445823073387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04199089854955673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.03893505781888962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.03818897530436516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.02450370043516159,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.020580368116497993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02021637000143528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.018184015527367592,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.017703499644994736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.013272203505039215,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01354095060378313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.012265851721167564,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.009716490283608437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.04273678734898567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.03584158048033714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.03184032440185547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.028262462466955185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.01939507946372032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.015999650582671165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.025632822886109352,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.02331860177218914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.020244425162672997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.01563546620309353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.015160813927650452,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.013055362738668919,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.011152705177664757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.009398764930665493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.008941682986915112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.006538650952279568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.004987573716789484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.004653621930629015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.004263903480023146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.003960269037634134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.003473845310509205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0034712685737758875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.002868041628971696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0023902307730168104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0428481288254261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.03501201048493385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.029858678579330444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.026616744697093964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.019089410081505775,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.01491191703826189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.02664991095662117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.024140529334545135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.020131420344114304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.015170253813266754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.014985787682235241,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.01351439580321312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.011503629386425018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.00922862533479929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.008605791255831718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.006766557227820158,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0048356736078858376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0043977960012853146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.004073855467140675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0036462252028286457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0035126814618706703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0034059712197631598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0026915359776467085,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.002188342157751322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1756085604429245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.14955584704875946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.13890667259693146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.12239907681941986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08067318052053452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.06992373615503311,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09721731394529343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08924295753240585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0834602639079094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.06486669182777405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06175591051578522,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.04936389625072479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04259105026721954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.03869667276740074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03773403912782669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02467101626098156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01980268396437168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.018941247835755348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.016586411744356155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.015916477888822556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012710344977676868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012334806844592094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011317485012114048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007636150810867548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.17265617847442627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.15349774062633514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.14630544185638428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.125789612531662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08070380240678787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07379523664712906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09464309364557266,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0860721543431282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08211905509233475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0660165399312973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06170710548758507,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0487380288541317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.041771307587623596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.039406705647706985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.038827069103717804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024621980264782906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021481908857822418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02098594233393669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018773801624774933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018404876813292503,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013682577759027481,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014801764860749245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012944573536515236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011539943516254425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.209043487906456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.19700421392917633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.19321338832378387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.17542071640491486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09851204603910446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09449568390846252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1090313121676445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10056618601083755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0993163213133812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08842168003320694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08447795361280441,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05569034069776535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04819665849208832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04729839414358139,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04708410054445267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02784634381532669,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.024610033258795738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.024359652772545815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02261015959084034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.022482460364699364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.014900386333465576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015216012485325336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.014612419530749321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010649471543729305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24484506249427795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23094336688518524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22673256695270538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.2060040980577469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11548089236021042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11090388149023056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1278538554906845,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11774861067533493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11639799177646637,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1037989929318428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09926366060972214,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0653105154633522,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05634693056344986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05533489212393761,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05509016662836075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032666780054569244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028497371822595596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.028219513595104218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026142287999391556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025994718074798584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0174561757594347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017186559736728668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01712212525308132,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011524848639965057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.3.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20301590859889984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.17993180453777313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17123068869113922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15232975780963898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09230129420757294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08366638422012329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.10861063003540039,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.09921830892562866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09496082365512848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07847336679697037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07448621094226837,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05513899773359299,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04751267284154892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.044402044266462326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04363902285695076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.027674056589603424,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.023362280800938606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.022996846586465836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.020694028586149216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020208235830068588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.014936521649360657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015182916074991226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.013920888304710388,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.010784035548567772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.05948399379849434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05045810714364052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0450628362596035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04024699702858925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.02709878981113434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.022555913776159286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.03599774092435837,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03251658007502556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.02822723612189293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.022175561636686325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02156623639166355,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.01832883432507515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.015611917711794376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.013163863681256771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.012529414147138596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.009210369549691677,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.007052582688629627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.006607340648770332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.006121970247477293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.00571369007229805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.004913266748189926,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.004976646043360233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0040607331320643425,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003532960545271635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.05663783848285675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.04680328816175461,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04078899696469307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.036554399877786636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.025308020412921906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.020324628800153732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03483179956674576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.031312715262174606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.026685133576393127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02047596499323845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.020128708332777023,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.01776907965540886,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.014995941892266273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01226199883967638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.011548623442649841,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.008920381776988506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.006504755932837725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0060025774873793125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.005570911802351475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005091105587780476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.004694020375609398,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0045917523093521595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0037039704620838165,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.003107720520347357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1917179524898529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16577664017677307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15542399883270264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1376524418592453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08840838074684143,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07796114683151245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1050894558429718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09659653156995773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09124546498060226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07238545268774033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06868880242109299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05346288904547691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.046208061277866364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04242615029215813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.041495393961668015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.026723310351371765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.021805057302117348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02098340354859829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01855577528476715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0179296862334013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013873790390789509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013583481311798096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012519973330199718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008629865944385529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.16939057409763336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1513446867465973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.14464770257472992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.12346041202545166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.079004667699337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07215382903814316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.09365367144346237,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.08410616219043732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08046706020832062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06545904278755188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0601811446249485,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04818735271692276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.040927812457084656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.03869343549013138,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.038167115300893784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.024444689974188805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021173864603042603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.020702725276350975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018659424036741257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018337540328502655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013722981326282024,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01465285662561655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013013369403779507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011514890007674694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1883140653371811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.17655447125434875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1725977212190628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.15641066431999207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0888027548789978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.08470439165830612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09940768033266068,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0913398340344429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08968131244182587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07924304902553558,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07571203261613846,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05076218023896217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04382229968905449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.042683910578489304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.042410291731357574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.025482339784502983,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02221999131143093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02194603905081749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.020312026143074036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.020150791853666306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.013768038712441921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013788512907922268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.013400079682469368,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009614584967494011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.246357262134552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23156091570854187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22673998773097992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20575161278247833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11632581055164337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11125487089157104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12949900329113007,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11917395889759064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11740680038928986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10403173416852951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09944429248571396,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06603679060935974,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05707656964659691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05578989163041115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05549166351556778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03308970108628273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028737511485815048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02841159887611866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026251301169395447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02606653794646263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0176161490380764,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01735866442322731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.017196768894791603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011594372801482677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.4.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2095576524734497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18695861101150513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17822471261024475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15851232409477234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09566449373960495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08714070916175842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11320783942937851,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10278883576393127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09833946079015732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08187927305698395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0775674432516098,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.0575646311044693,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04935459420084953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046164318919181824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04538174346089363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028994236141443253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024500075727701187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024127205833792686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021858954802155495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02137257158756256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015817200765013695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016173504292964935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014756884425878525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011794302612543106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07179942727088928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.06272782385349274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0575268529355526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05125272274017334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03292565792798996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.028575770556926727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04211081191897392,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03832326829433441,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03401795029640198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.027597596868872643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02656625211238861,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.021354198455810547,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01834231987595558,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01593482308089733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.015323616564273834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010706116445362568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008427944034337997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007987068966031075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007394589949399233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.006993317045271397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.005677962210029364,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005734641570597887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004852258134633303,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003964860457926989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.06473594158887863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.055526718497276306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.049911461770534515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.044354259967803955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.02931882254779339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.02469693124294281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03860938921570778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03517904505133629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.030482374131679535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.024252815172076225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023485815152525902,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019544633105397224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016757169738411903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.014120960608124733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01342977024614811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009770752862095833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0073308334685862064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006842531729489565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006326897535473108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005861466750502586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0050523970276117325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004929880145937204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004123569931834936,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.003137608990073204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.20667532086372375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.18412651121616364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.17575180530548096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1560567021369934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09606868773698807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08750919252634048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.11123546212911606,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10240286588668823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09829515218734741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08090247958898544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07650333642959595,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.056504372507333755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04887129366397858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.046007949858903885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.045349981635808945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0281996950507164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.023354772478342056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.022706395015120506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02021184004843235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.019752709195017815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014485612511634827,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013975773938000202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013443267904222012,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008471165783703327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18959152698516846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17195124924182892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1653534173965454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14394891262054443,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08860822021961212,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0817338302731514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10249451547861099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0936884954571724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09002437442541122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0746944472193718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07026924192905426,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05280518904328346,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.045185159891843796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04280707612633705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04224920645356178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026574712246656418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02261975221335888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022079620510339737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.019902443513274193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.019540734589099884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01446588896214962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014702496118843555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013650595210492611,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010650047101080418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1762571483850479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1653290092945099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.16162365674972534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14647088944911957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08305193483829498,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07925103604793549,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09263620525598526,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08550563454627991,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08388134837150574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07411837577819824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07074187695980072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.047173138707876205,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.040913499891757965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.039821311831474304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03956005722284317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02358744852244854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020505355671048164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020246082916855812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018697958439588547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018533287569880486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012398071587085724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012418893165886402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012031331658363342,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00827084295451641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2419440895318985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2276521474123001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2230045050382614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20235870778560638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1143060028553009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10936648398637772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12692731618881226,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11716075241565704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11531758308410645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10232201963663101,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09775715321302414,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06462647765874863,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.056016407907009125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05473296344280243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05443199723958969,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032305825501680374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02803993970155716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027713479474186897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02559611015021801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025405537337064743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016914283856749535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016712086275219917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01648643985390663,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010864359326660633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.5.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2036168873310089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1832730621099472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17470474541187286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1561361700296402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0934390053153038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08529166877269745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.111430324614048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10100080072879791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09583073854446411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08062601834535599,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07705391943454742,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05686961114406586,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04845084622502327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04500170052051544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04415525868535042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028722483664751053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02372308075428009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023303192108869553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021242249757051468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020712487399578094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015683121979236603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01551529485732317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014488616958260536,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011047808453440666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.06634723395109177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.05762307345867157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.051811300218105316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.04625110328197479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.030352430418133736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.025681382045149803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04047377035021782,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.03665715083479881,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03144051507115364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.025375576689839363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.02464967966079712,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.020561281591653824,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.017536645755171776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.014713982120156288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.013981547206640244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010300645604729652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.007822458632290363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007315567694604397,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.006863356567919254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.006390023976564407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0054395729675889015,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005470580421388149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004440734162926674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.003781738691031933,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.060605134814977646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.051686182618141174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04550164192914963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.04061912000179291,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.02736980840563774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.022464901208877563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03753753378987312,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.034015703946352005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.02854038216173649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.022614534944295883,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.022116539999842644,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019017454236745834,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016205081716179848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013237089850008488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.012449914589524269,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009491239674389362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.006952513940632343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006399805191904306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006020466797053814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.005488865077495575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0049445740878582,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004851828329265118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.003889617044478655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0031672290060669184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.18390421569347382,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16390341520309448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1555459350347519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.13815124332904816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0854179784655571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07736478745937347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10099665075540543,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09283767640590668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08745920658111572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07208962738513947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06840716302394867,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.051350172609090805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04431234672665596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04098079353570938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.040139734745025635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.02561933919787407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020926428958773613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.020221518352627754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01816825568675995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017614122480154037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013170411810278893,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012800670228898525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011958121322095394,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007893593981862068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18748553097248077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1659504473209381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15739239752292633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.13692741096019745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08746771514415741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07871513068675995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10398761928081512,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09420737624168396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0894230529665947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07223328202962875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06758613139390945,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.053499579429626465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.045534636825323105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04249802604317665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.041758015751838684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026858501136302948,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02269449643790722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02203347347676754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01975790224969387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.019285235553979874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0145133500918746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015179971233010292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01350901834666729,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011253130622208118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17114149034023285,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.16083428263664246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15720336139202118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14284665882587433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08073896169662476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07706644386053085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09042980521917343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08328168839216232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08152130991220474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07229864597320557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06913543492555618,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046180956065654755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039897676557302475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03875955566763878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03848888725042343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02312663570046425,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020065199583768845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019804779440164566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018367337062954903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018200399354100227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012346301227807999,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012314205057919025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011960572563111782,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008389642462134361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23557148873806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22183120250701904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21720033884048462,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19756180047988892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11134614795446396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10655659437179565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12412071228027344,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11433691531419754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11234697699546814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09990664571523666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0955205112695694,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06329933553934097,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054706476628780365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05334842950105667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05303625017404556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03165601193904877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02740606665611267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027081916108727455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02507716789841652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024877533316612244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016712261363863945,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016471730545163155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016263943165540695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010853748768568039,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.6.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21076911687850952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19050660729408264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18252255022525787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16333793103694916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09705882519483566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08929018676280975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11463408172130585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10375270992517471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09938328713178635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08413700014352798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08029516786336899,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05836555361747742,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.04982609674334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04680825024843216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.046071410179138184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029368113726377487,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024785494431853294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024426598101854324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02229994349181652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02183464542031288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015965718775987625,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016251832246780396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014962779358029366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011789047159254551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07253632694482803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.06478405743837357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05837472155690193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05205954238772392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03348775953054428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.02869037352502346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04561757668852806,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04102073982357979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03436758741736412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.028703801333904266,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.027966883033514023,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.023239726200699806,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01973414234817028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01628398336470127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.015371035784482956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.011651421897113323,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008739606477320194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.008151110261678696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007829796522855759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.007258750032633543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.006187691818922758,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.006257256492972374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.004955670330673456,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.004403706174343824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0626121312379837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.05524720624089241,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.04855867102742195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.043166615068912506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.028583906590938568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.023709077388048172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04002339765429497,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03640666976571083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.029457036405801773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02430696412920952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023732928559184074,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02031564898788929,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.017455879598855972,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013826129958033562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.012836582958698273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.010184375569224358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.007291768211871386,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006658272352069616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006462015211582184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0058193085715174675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.005293699912726879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.005209335591644049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.00403254572302103,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0033859771210700274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19899427890777588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17977093160152435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1722375452518463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.15291020274162292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09298620373010635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0854603573679924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10868975520133972,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09914804995059967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09474433958530426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07913115620613098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07504456490278244,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.055301252752542496,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04738381505012512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04455778747797012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04388604313135147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027598528191447258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022764069959521294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0221426859498024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019917359575629234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.019456520676612854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014231402426958084,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01377798430621624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013111609034240246,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008610040880739689,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1884087473154068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.16933833062648773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16095305979251862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1393108069896698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08832570910453796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08029365539550781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10636451095342636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09553088247776031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09012182801961899,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07366209477186203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0699334368109703,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.054981034249067307,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.046462249010801315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04323485121130943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04246465116739273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027733784168958664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023538334295153618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022891977801918983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.020789284259080887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020309122279286385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015384198166429996,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01624443754553795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.014347760006785393,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012545468285679817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1666836440563202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1567574441432953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1534167230129242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13940130174160004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0788261890411377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07530619204044342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08801373839378357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08102373778820038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07955755293369293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07058826088905334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06748536229133606,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04497779533267021,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03889338672161102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0379004031419754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03766704350709915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02252550795674324,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01975112594664097,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019513443112373352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01810702309012413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0179611723870039,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012069562450051308,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012280250899493694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01174020767211914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008592424914240837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23579122126102448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.222219780087471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21776892244815826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1981118619441986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1117170974612236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10700331628322601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12449662387371063,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11445385217666626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11269330978393555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10021616518497467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09587167948484421,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06365256011486053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05487082898616791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053646303713321686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.053360287100076675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03187926858663559,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027765214443206787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0274557713419199,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025439215824007988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02525932528078556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01705533266067505,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016955142840743065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016653813421726227,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.0115723367780447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.7.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20433014631271362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18405292928218842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1758449673652649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15770600736141205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09402231127023697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08595852553844452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11139664053916931,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10105358064174652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09641910344362259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08123823255300522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07780414819717407,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.056854426860809326,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.048471637070178986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04529405012726784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044517822563648224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.028560878708958626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02386007271707058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023480109870433807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021384425461292267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020891698077321053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.015477100387215614,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015538507141172886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014423436485230923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011068885214626789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.07229823619127274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0640784278512001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.05915708839893341,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05233281850814819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.03329877555370331,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.029231058433651924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.042107172310352325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.038525331765413284,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03417966887354851,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.028040915727615356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.026857197284698486,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0213544350117445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.01839970052242279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.01605963334441185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.01547007542103529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.010699591599404812,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.008419157937169075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.007991536520421505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.007386090699583292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.00699745723977685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.005629621911793947,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.005631897132843733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0048264265060424805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0037945793010294437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.06344876438379288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.05514335259795189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.05007714778184891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.044085338711738586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.028834931552410126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.024654246866703033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.03750180825591087,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.034161586314439774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.029770636931061745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.023919442668557167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.023006537929177284,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.019016969949007034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.016308337450027466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.013897260650992393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.013273438438773155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.009519466198980808,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.0072477711364626884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.006803502328693867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.006268806755542755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.0058531141839921474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0049530318938195705,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.004873728379607201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004104185849428177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0031792030204087496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.18240326642990112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1626574695110321,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15496958792209625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.13663387298583984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08455818891525269,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07684142887592316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09996659308671951,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09083674848079681,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08648092299699783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07106637209653854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06717733293771744,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05087890848517418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04350433498620987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04059627652168274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.039900291711091995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.025416448712348938,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020873580127954483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.020246481522917747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.018127888441085815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017663544043898582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013154354877769947,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012906396761536598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012026366777718067,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008315572515130043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1977970451116562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18005962669849396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17297670245170593,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15205296874046326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09264685213565826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0856722965836525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10854403674602509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09851948916912079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09423834830522537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07904263585805893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07453982532024384,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05574023351073265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04742798954248428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.044760387390851974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.044106315821409225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027905823662877083,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023507753387093544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022936217486858368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02082017809152603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020411867648363113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014840735122561455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015165221877396107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013953466899693012,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010774858295917511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16268427670001984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1532304435968399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14993759989738464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1363285779953003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07694627344608307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0735417902469635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08607260882854462,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07920283079147339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07766101509332657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06899941712617874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0660800039768219,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0439998134970665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.037978317588567734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.036970168352127075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03672965243458748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02201676554977894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01919376291334629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018961550667881966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01760033145546913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017453495413064957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011735353618860245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011853777803480625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011399183422327042,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008189890533685684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23203247785568237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21882356703281403,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2143489271402359,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1949910670518875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10994546860456467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10524184256792068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1226196438074112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11273852735757828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11088178306818008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09857067465782166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09438752382993698,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06263895332813263,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05398797243833542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.052726712077856064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.052423760294914246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03131991624832153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027149271219968796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026830697432160378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024843666702508926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02466162107884884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016555219888687134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016400840133428574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01612219400703907,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010934562422335148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.8.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21018268167972565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19030331075191498,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18263329565525055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16296975314617157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09686324745416641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0892181545495987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11363700777292252,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10320858657360077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09913817793130875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08380107581615448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0797918289899826,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05793079733848572,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049579475075006485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046735674142837524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04604290425777435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029255900532007217,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024781428277492523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02444753423333168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022280484437942505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021841231733560562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016100801527500153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016270168125629425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015168504789471626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.01186355110257864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.0923733115196228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08220721781253815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.07622025161981583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.06749071180820465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04258981719613075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.037622444331645966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05377095192670822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04885927587747574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04363660514354706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0360131561756134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03449761122465134,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02736286073923111,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.023438431322574615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02063322439789772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.01992359384894371,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.013733101077377796,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.010986043140292168,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010486569255590439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.009727832861244678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009283306077122688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.007337464485317469,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007517325226217508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00634734658524394,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00537061644718051,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.07935116440057755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.06979311257600784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06384820491075516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05638270825147629,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03625747188925743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.031399913132190704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04671098664402962,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04258308187127113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.0373074896633625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.030413977801799774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.02923060953617096,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.023759830743074417,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020337557420134544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.017467139288783073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01673617959022522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.011891154572367668,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009111745283007622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008590858429670334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.007955510169267654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007472165394574404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00619489885866642,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006100549828261137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.0051757474429905415,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00400660140439868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19220881164073944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17436790466308594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.16780315339565277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14817847311496735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0897936075925827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08307909220457077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10316525399684906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09467410296201706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09124304354190826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07631740719079971,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07175160944461823,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.052423689514398575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.045171648263931274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04294484108686447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04242084175348282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.026141708716750145,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.021819069981575012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02130700834095478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019036483019590378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01867504231631756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013404963538050652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012975042685866356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012573492713272572,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007934489287436008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1993076205253601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18014933168888092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17243245244026184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15134650468826294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09344325959682465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08576633036136627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10941915959119797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09956937283277512,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09505053609609604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07860640436410904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07455437630414963,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05619671940803528,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04785337299108505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04504535347223282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04437866061925888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.028088334947824478,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02348434552550316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022888371720910072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02051377296447754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020069656893610954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014821244403719902,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014906519092619419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013882310129702091,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01027657650411129,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16206996142864227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15213730931282043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14868398010730743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13507655262947083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07658286392688751,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07301879674196243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08578245341777802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07891295105218887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07731891423463821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0684279277920723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06548789143562317,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.043890975415706635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03789806365966797,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.036841198801994324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03659507632255554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021988868713378906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019250184297561646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019006384536623955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017626376822590828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017475441098213196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0117854755371809,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01205415092408657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011435518972575665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008514638058841228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22811272740364075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21440167725086212,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2097819298505783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19063790142536163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10790391266345978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10301275551319122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12053030729293823,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11082379519939423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10892365127801895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.096433125436306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09228646010160446,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06161956861615181,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053128574043512344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.051801811903715134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05148930847644806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030858976766467094,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026810333132743835,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026479406282305717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024499699473381042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024301273748278618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016448548063635826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016406487673521042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016002364456653595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011173042468726635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.9.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21688273549079895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.195871502161026,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18770845234394073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16686895489692688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09991750866174698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09185732901096344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11755452305078506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10667914897203445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10227162390947342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08605503290891647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0815466120839119,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05971937254071236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05112798139452934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.048075463622808456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04733084887266159,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030013510957360268,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02524554915726185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.024882636964321136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022554337978363037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022082101553678513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016292694956064224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016276968643069267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015285375528037548,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011505048722028732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.08111102879047394,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.07254783809185028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.06766746938228607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.05942074581980705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.037567462772130966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.03346123546361923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.04657381772994995,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.042521990835666656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.03834857791662216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03160262852907181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.030077604576945305,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.023668518289923668,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.020309679210186005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.018092622980475426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.017540713772177696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.011844425462186337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.00947027001529932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.009059125557541847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.008297783322632313,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.007934422232210636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.006231253035366535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.006245048716664314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00544535368680954,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.004236916080117226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.07359469681978226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.06516092270612717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06042276322841644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05277974158525467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.033782899379730225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.029807444661855698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.042435433715581894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.03851376846432686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.034619253128767014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.02816605381667614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.026803895831108093,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02149576134979725,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.018389325588941574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.016235601156949997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01569240540266037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.010735702700912952,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.008401831611990929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.007987060584127903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.007274224888533354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.00690606702119112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0055682044476270676,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.005442801862955093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.004773732740432024,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0034874265547841787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.17976276576519012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.16189919412136078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.15510153770446777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1357804536819458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08386819809675217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.07699833065271378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.09768102318048477,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.08885928243398666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.08545619249343872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07042080163955688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06619781255722046,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.049674712121486664,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04243537038564682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04017745330929756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.03963658586144447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.024822253733873367,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.020555876195430756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02002415992319584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.017803162336349487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.017433637753129005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.012822971679270267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.012464363127946854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.011917990632355213,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.007906206883490086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19702740013599396,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17771965265274048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16960251331329346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1487310379743576,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09191865473985672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08419815450906754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10898296535015106,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09889288991689682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09361158311367035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0775689110159874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07357574999332428,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.056102555245161057,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04769488424062729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04453796520829201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04377840459346771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.028172001242637634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.023658720776438713,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02303505875170231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.0208913441747427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020396705716848373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015303699299693108,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01567230373620987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01426894124597311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01145181804895401,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16228733956813812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15147458016872406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.147598534822464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.133662149310112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07651983946561813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07260174304246902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08628086000680923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07926963269710541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07741113752126694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06795590370893478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0649128332734108,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04409165307879448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03805132955312729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03680633008480072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03651457279920578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022104473784565926,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019228477030992508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018950073048472404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01753089763224125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0173508133739233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011845353990793228,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012083268724381924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011417168192565441,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00851461011916399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22261309623718262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.20802195370197296,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20303837954998016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18380849063396454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1050167977809906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0998120978474617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11781039834022522,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10840636491775513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10619650781154633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0932191014289856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0889882817864418,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06010079383850098,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0518934428691864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.050366226583719254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.050000064074993134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030053550377488136,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.025963500142097473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02559644542634487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02357950061559677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023345347493886948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.015806002542376518,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015786699950695038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015273896045982838,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010550056584179401,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.10.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22963909804821014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20756880939006805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19922778010368347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1766587495803833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1061212494969368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0978289544582367,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12449796497821808,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11283789575099945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10863349586725235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09140748530626297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08657213300466537,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06365633755922318,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0542290173470974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.051196929067373276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05046088993549347,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03204844519495964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027043066918849945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026687270030379295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02419471926987171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023724334314465523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017471883445978165,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017611542716622353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01644137129187584,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012711179442703724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09471038728952408,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08441315591335297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.07877811789512634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.06901639699935913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04399825632572174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.03916407749056816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05446501821279526,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04960593208670616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04493807256221771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03683320805430412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03501424565911293,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02770739234983921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.023780781775712967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.021279659122228622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.020652614533901215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.013886299915611744,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.011258203536272049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010791515931487083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.009872546419501305,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009467078372836113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0073661478236317635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007558855228126049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006490133237093687,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005331674125045538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0793822705745697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07023479789495468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06453816592693329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05635121092200279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03643607348203659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03185543790459633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04654272645711899,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04240185767412186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.03729833662509918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03040153533220291,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.02904202975332737,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.023674670606851578,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020249824970960617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01757160946726799,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.016896599903702736,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.011824156157672405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009196280501782894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008705075830221176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00800666119903326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.007557692006230354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00619993731379509,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006152571178972721,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005265026353299618,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004109062720090151,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.1938401460647583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17283737659454346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1647641360759735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14326342940330505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09012022614479065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08195556700229645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10631053894758224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0960053950548172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09189220517873764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07470855861902237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.06997879594564438,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054187141358852386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04590385779738426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04318470135331154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04252967983484268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027100728824734688,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022095059975981712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021475177258253098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01895921491086483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.018507080152630806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013989497907459736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.013476541265845299,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012775829993188381,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008517906069755554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19575975835323334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17775772511959076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16314177215099335,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14397133886814117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09221747517585754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08105906844139099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12340552359819412,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10978596657514572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0938209667801857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0790930762887001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0766202062368393,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06391671299934387,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05313868075609207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04506020247936249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04295142740011215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032097749412059784,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.024306951090693474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.022937802597880363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02180151827633381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020510662347078323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017426537349820137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01731007546186447,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0146400835365057,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012545672245323658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1656651645898819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15426109731197357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15001104772090912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13573534786701202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.078199602663517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07397285103797913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08861343562602997,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08129531145095825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07914450764656067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06920995563268661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06611698120832443,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04535221680998802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03911295160651207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03772561997175217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03739975392818451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022767936810851097,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019898805767297745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01959347166121006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01814555749297142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017941074445843697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012317722663283348,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012772978283464909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011842994019389153,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009278111159801483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2246875762939453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2094203233718872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20410379767417908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18452906608581543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10587392002344131,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10035202652215958,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1192048043012619,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10953209549188614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1070624440908432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09369143098592758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0893295407295227,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06066050007939339,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05242482200264931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05077127739787102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05038008093833923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030338531360030174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026141522452235222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025747839361429214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02367529831826687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023426810279488564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.0158806461840868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.015857579186558723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01529465802013874,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010516893118619919,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.11.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22915378212928772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20730051398277283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19904541969299316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17664343118667603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10618459433317184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09784796833992004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12465621531009674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1128915399312973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10870573669672012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09153155237436295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08664526045322418,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06376823782920837,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05429444834589958,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05123336985707283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05048857629299164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03208930417895317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027049902826547623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026694169268012047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02420920506119728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023741841316223145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017339862883090973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017597826197743416,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.016272757202386856,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012688130140304565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09936142712831497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0882679894566536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08140312135219574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07137010246515274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04610950127243996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04047267511487007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.059583116322755814,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05331159383058548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04718722775578499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03861810266971588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03724299371242523,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.030456092208623886,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.025676630437374115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02238871343433857,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.021564612165093422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.015369811095297337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0120000084862113,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011409792117774487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010573736391961575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010039178654551506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.008258509449660778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008328622207045555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007047239691019058,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0059978412464261055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08464791625738144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07457271218299866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.0683441236615181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.05965745449066162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.0388471893966198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03373681753873825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.050359148532152176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04534270614385605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.039873044937849045,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03232743963599205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.030980568379163742,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025655295699834824,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0217230673879385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.01878642849624157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018025169149041176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012857151217758656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009926670230925083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009392314590513706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008654155768454075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008158053271472454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006775897927582264,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006758794654160738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005715447477996349,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004645415581762791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19185824692249298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.17256228625774384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.16470789909362793,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.14446885883808136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.08987101912498474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08207979798316956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.10619068145751953,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.09596749395132065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09148573130369186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.07533403486013412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07105202227830887,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.054195694625377655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04592197388410568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04309283569455147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04241381585597992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.027088407427072525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02205561101436615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021414682269096375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.019084369763731956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01862308569252491,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.013995630666613579,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01345471478998661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012836312875151634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008503880351781845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.20962904393672943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18562078475952148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17620067298412323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15445590019226074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09794670343399048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08820542693138123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.116356261074543,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10519334673881531,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10026030987501144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08127445727586746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07676979154348373,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06004529073834419,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05087818577885628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04757343977689743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04677799716591835,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03023526445031166,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.025310833007097244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.024546710774302483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.022090625017881393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.021575355902314186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01639850251376629,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.0168004110455513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01531671267002821,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012325716204941273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16898709535598755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1571710705757141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15290850400924683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1381736844778061,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07989440113306046,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07549551129341125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09060564637184143,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08289023488759995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08087866753339767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07055147737264633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06759300827980042,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04641929268836975,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039985645562410355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.038659267127513885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03834491968154907,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02337837591767311,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02056017331779003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020253784954547882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018753837794065475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018560145050287247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012866825796663761,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013408723287284374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01244029775261879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009980638511478901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23236511647701263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21642330288887024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2109871655702591,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1904449611902237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10978308320045471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10397175699472427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12320615351200104,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11323282122612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11103887110948563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09688219428062439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09235727041959763,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06292006373405457,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.054283592849969864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.052717555314302444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05233968421816826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03146981820464134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02724529057741165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0268524382263422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024647196754813194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02441091276705265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016622863709926605,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016672734171152115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01608199253678322,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01129305362701416,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.12.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.23011666536331177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.2070072591304779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19730734825134277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17449317872524261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1063273698091507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09702161699533463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12708328664302826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11497606337070465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10899266600608826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09107682853937149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08619516342878342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06467711925506592,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05527600273489952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05132497847080231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05035177245736122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03264697268605232,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02715372107923031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02667497657239437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.0242130346596241,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023597614839673042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017836224287748337,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017873873934149742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01653091423213482,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012847579084336758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10213909298181534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09181688725948334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08669238537549973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07614187151193619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04754956066608429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04299907013773918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.057521212846040726,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05237749591469765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0484396331012249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.040061332285404205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03804078698158264,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.029256463050842285,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02503294311463833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0228714719414711,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022345155477523804,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01463894173502922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0118982819840312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011480354703962803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010407092981040478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.01006257627159357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.007670809514820576,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00764026353135705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006871089804917574,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005136272870004177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08364420384168625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07518991827964783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07015972584486008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06166722998023033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03878398612141609,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03463580459356308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.04810282588005066,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.043912373483181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.039543092250823975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.032745540142059326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.031173640862107277,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.024396320804953575,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.020994562655687332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018659228459000587,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018081730231642723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012200326658785343,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009689881466329098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.0092628113925457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008484462276101112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008097478188574314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006369702983647585,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006301886402070522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.00556590873748064,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004146175924688578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.20933467149734497,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.1882956326007843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18069924414157867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.158315509557724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09775286912918091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08979347348213196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1138065904378891,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10322064906358719,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09948767721652985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08194328844547272,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07698874920606613,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.05798124149441719,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04929068312048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04681167006492615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04621042311191559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.028967253863811493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02381446212530136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02323269098997116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02055460587143898,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02015010267496109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.014868661761283875,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.014231311157345772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013744518160820007,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.008764226920902729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.22750461101531982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.2072417140007019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19912861287593842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17833483219146729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10650797188282013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09856891632080078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12446413189172745,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11342708766460419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10829344391822815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09154056757688522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08819243311882019,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06406870484352112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0546913668513298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05158976837992668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.050834186375141144,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032115962356328964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027341658249497414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026722058653831482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.024411490187048912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.023940661922097206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01732906885445118,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.017942391335964203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016332853585481644,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013093994930386543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17158488929271698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1596268117427826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15523813664913177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14031267166137695,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08117655664682388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07666884362697601,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09198509156703949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08441497385501862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08218464255332947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07170307636260986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06856146454811096,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04721067100763321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04066522419452667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.039182402193546295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03883276879787445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02365320362150669,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020666969940066338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020345376804471016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01881992071866989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018607188016176224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012764004059135914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013277272693812847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012260274961590767,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009646973572671413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23705653846263885,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22088705003261566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2152370661497116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19453756511211395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1120653972029686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10619194060564041,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1260979324579239,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11580414324998856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11338284611701965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09901244193315506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0943223088979721,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06442520767450333,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.055486634373664856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05379866436123848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05340783670544624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032245948910713196,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02780304290354252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02740365080535412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025173520669341087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024921493604779243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017092501744627953,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01700826920568943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016501493752002716,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011473393999040127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.13.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2439960241317749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.21875041723251343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.20916424691677094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.18457379937171936,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1127302423119545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.1031016856431961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.13287678360939026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1204565018415451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.11572746932506561,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09608486294746399,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0907578244805336,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06803296506404877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05786415562033653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05433708801865578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.05347524583339691,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03423671796917915,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.0285765640437603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.0281657837331295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02533121034502983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.0247786957770586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.018537364900112152,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01849132962524891,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.017353331670165062,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013140566647052765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10424488037824631,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09404675662517548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08792061358690262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07763238996267319,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.048578958958387375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.043544284999370575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06060922145843506,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0549757182598114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04951776936650276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04127885401248932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03949272260069847,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.030869731679558754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02632269635796547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.023477280512452126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022763244807720184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.015449536964297295,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012360786087810993,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011845728382468224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.010933089070022106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010474280454218388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00815539713948965,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00824003480374813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0071189384907484055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005706408992409706,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0841318741440773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.07603752613067627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.0695098266005516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.061464838683605194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.038894351571798325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03407863900065422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.051064424216747284,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04636864736676216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.03965947777032852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.033343810588121414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03209308907389641,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025806216523051262,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.022154830396175385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018763301894068718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.017897984012961388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012900080531835556,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009814596734941006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00923085305839777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00871579721570015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008151985704898834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006758652627468109,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00665030675008893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005608730483800173,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004370537586510181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2194114476442337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19817328453063965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1897161304950714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1665915846824646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10261544585227966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09422661364078522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12187033146619797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10919661074876785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10440018773078918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08657117933034897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08166251331567764,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06234155595302582,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05223219096660614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04920317605137825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04848750680685043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.0311729833483696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025239553302526474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024584274739027023,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.021982988342642784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021497495472431183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016114113852381706,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015443004667758942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014659930020570755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009926020167768002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2451043426990509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.21659700572490692,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.20483578741550446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17874456942081451,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.11447533965110779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.10334701091051102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.1379907876253128,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.12453643232584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.11774825304746628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09409315139055252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0903477817773819,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.07104182243347168,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.06020486354827881,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.055550310760736465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.054439641535282135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03559456765651703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.029506895691156387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02864748425781727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.025540543720126152,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02481147274374962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.019040733575820923,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.019608929753303528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01755474880337715,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.014199887402355671,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1612621545791626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15026816725730896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14586375653743744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13207511603832245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07646261900663376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07212352752685547,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08729451894760132,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08010675758123398,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07737758010625839,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06777670234441757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06485375761985779,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044831451028585434,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03866015747189522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037003252655267715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03661582991480827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022499021142721176,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019655419513583183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019326208159327507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017969045788049698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01773681491613388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012186964973807335,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01285128016024828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011632228270173073,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.009495903737843037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2333511859178543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2181156873703003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21250203251838684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19235879182815552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11058540642261505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10485655814409256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12495282292366028,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11465955525636673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1117902547121048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09811998158693314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09367595613002777,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06389470398426056,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.055031534284353256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05319531261920929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05276918783783913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0320056714117527,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02765602432191372,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02723834291100502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02515741065144539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024881061166524887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01706412434577942,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017189128324389458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016435060650110245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01187122892588377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.14.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22905723750591278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20517916977405548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19521762430667877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1729503870010376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10571648925542831,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09612210839986801,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12685558199882507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11451590061187744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.1084742546081543,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09032357484102249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08582853525876999,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06501206755638123,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05518336594104767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0513480007648468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.0503263846039772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03308285027742386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02759529836475849,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.027087146416306496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.024699050933122635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.024051770567893982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01840297505259514,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.018716109916567802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01693183183670044,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013982602395117283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11666414886713028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10602928698062897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09970837086439133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08854666352272034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05452314764261246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.049299050122499466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06797266751527786,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06117452681064606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.055459439754486084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04683902859687805,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04488718509674072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03467332944273949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02940622717142105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.026408664882183075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025664405897259712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0174703486263752,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014032304286956787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013494862243533134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01255026925355196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012077713385224342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009334007278084755,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009463821537792683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008200091309845448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006772384513169527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08810259401798248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08004835993051529,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07284713536500931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06478346884250641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04067221283912659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03553685545921326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05450048670172691,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04918598383665085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.041490521281957626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03524171561002731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03419729322195053,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.027716979384422302,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.023541107773780823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.019659943878650665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01862768456339836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.013893683440983295,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010324222035706043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00965107511729002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009246932342648506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008591270074248314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007267958018928766,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007098592352122068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005894139409065247,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004685773514211178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23456385731697083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21531420946121216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.20743320882320404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1857202649116516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11070002615451813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10285980999469757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1297006905078888,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1173039972782135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11238881200551987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09601468592882156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09145043045282364,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06627190858125687,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0561567060649395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05312134325504303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05239717289805412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033112697303295135,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027156980708241463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.026501482352614403,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02409745194017887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023611770942807198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01705765165388584,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01640031486749649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015800446271896362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01031299214810133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21899835765361786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1978422850370407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19002032279968262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16452579200267792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10246901959180832,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09439628571271896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11919937282800674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10794547945261002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10393045097589493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08561493456363678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07929619401693344,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06125788763165474,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0518881231546402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0493588000535965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.048741310834884644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03068169765174389,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02574397437274456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025154128670692444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02237648144364357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.021974915638566017,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016339421272277832,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01626567915081978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015499301254749298,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011294243857264519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15770180523395538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14742374420166016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14333368837833405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13005302846431732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.074795663356781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07083833962678909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08524028956890106,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07809461653232574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07567718625068665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06661991029977798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06384902447462082,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04373669996857643,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03773334249854088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03623216971755028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03588201850652695,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022020608186721802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01930447854101658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01900804601609707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017728151753544807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017513221129775047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012091686949133873,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012669697403907776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011600361205637455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00945026334375143,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23564356565475464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2211351990699768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21591977775096893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19587956368923187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11163628846406937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10627896338701248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12551924586296082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11538129299879074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11277883499860764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09955240041017532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09522084146738052,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06404978036880493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05523828789591789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053552087396383286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05315331742167473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03202453628182411,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027533670887351036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027150021865963936,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025073280557990074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024819286540150642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016862886026501656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016634929925203323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01628117449581623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010960489511489868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.15.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22647976875305176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.2024482637643814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1927223652601242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17141802608966827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10423047840595245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0946464091539383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12394973635673523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11255349963903427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.1070900559425354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08901431411504745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08450054377317429,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06320696324110031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05404765158891678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05026623606681824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04934534803032875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.031886156648397446,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02654283121228218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026087269186973572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023602349683642387,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023010283708572388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017442844808101654,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017400845885276794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01620781607925892,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012444248422980309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11590727418661118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10605204105377197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10048915445804596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08960055559873581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05413498356938362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.049528010189533234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06562458723783493,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.059830501675605774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05495759844779968,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04686499014496803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04476265236735344,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0334007665514946,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028617089614272118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.026046190410852432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025423923507332802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016733521595597267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01356329396367073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013093114830553532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012112772092223167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011715425178408623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00880417414009571,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008751925081014633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007870012894272804,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005899733863770962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09108318388462067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08345703780651093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07700448483228683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06886369735002518,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04217318817973137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.037533994764089584,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.054534588009119034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.049892932176589966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042882587760686874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.036832045763731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.035446710884571075,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02755543403327465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.023793211206793785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.020290987566113472,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019393084570765495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.013776133768260479,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010540287010371685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00992585439234972,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.00948261097073555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008879275992512703,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007188939023762941,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007032295223325491,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005993238650262356,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004482063464820385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23983266949653625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.22201016545295715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.21503150463104248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1933378279209137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11326155811548233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10614102333784103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13152951002120972,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11893036216497421,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11476211249828339,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09913219511508942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09454678744077682,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06720900535583496,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05683869868516922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05423944443464279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05360081046819687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03359147533774376,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027577079832553864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.027025625109672546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.024652138352394104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.024242153391242027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01724315993487835,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01637602411210537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.016067082062363625,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010034640319645405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2320849895477295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.20865845680236816,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.20059821009635925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1781105250120163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10818787664175034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.099416084587574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12471873313188553,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11383410543203354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.11023689061403275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.09156417101621628,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08662991970777512,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06421130895614624,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05503349006175995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05245098099112511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05181943252682686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0323946475982666,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027934659272432327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.027320686727762222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.024720456451177597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02433253638446331,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017744649201631546,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01840244047343731,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01691754162311554,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013658495619893074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14805495738983154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13856996595859528,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13454031944274902,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12218108773231506,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07016174495220184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06640314310789108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08033265918493271,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07361839711666107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0709780603647232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06259211897850037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06004016101360321,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04113738611340523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03550390899181366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03396492078900337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0335921049118042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.020667053759098053,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018040912225842476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01773693785071373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.016575664281845093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.0163434948772192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011241261847317219,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011804170906543732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010727626271545887,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008715817704796791,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22845053672790527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21470676362514496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20956233143806458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19028256833553314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1082024797797203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.1030900627374649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12179800122976303,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11202165484428406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10928066819906235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09672294557094574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09250153601169586,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06213308870792389,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05364133045077324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0519101582467556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.051498766988515854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.031056690961122513,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026702728122472763,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026321053504943848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024355579167604446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02410231903195381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016307581216096878,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016161657869815826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015705900266766548,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010651995427906513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.16.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21916551887989044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1945718377828598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1843707263469696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16384552419185638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1004827544093132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09058739244937897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12038393318653107,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10945025086402893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10355225205421448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08526989817619324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08110543340444565,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06147739291191101,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05259532108902931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04852879047393799,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04750606790184975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03106815181672573,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025667477399110794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02517629787325859,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022724425420165062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022072019055485725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016971994191408157,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01695694774389267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015614386647939682,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012133860029280186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.09702017903327942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.08892075717449188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08326403051614761,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0748206302523613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.045153599232435226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04079575836658478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.05668080598115921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0516507513821125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04597777873277664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.03949667513370514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03810432553291321,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.028803909197449684,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02470206469297409,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.021792028099298477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.021063625812530518,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.014429976232349873,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.011458257213234901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.010961330495774746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01034674234688282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.009877747856080532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.0075792958959937096,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.007655511610209942,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.006535821128636599,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005238090176135302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08191414177417755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0746629536151886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.06833002716302872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.061438750475645065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.03774355724453926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03322683274745941,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05006025731563568,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.04548916965723038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.038546204566955566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03305013105273247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03211662173271179,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.025281812995672226,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02165038511157036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.018216287717223167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01730922982096672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.012625445611774921,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.009527072310447693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.008940786123275757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.008605592884123325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008032230660319328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.006586558651179075,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.006493933964520693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005386063363403082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0042536910623312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.19983750581741333,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.18037481606006622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.17052434384822845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.15358620882034302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09235408902168274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08370831608772278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1147618442773819,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.10271640121936798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.09462103247642517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08016729354858398,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.07711547613143921,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.058539628982543945,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.04904281347990036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0443052276968956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04314000904560089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.029245130717754364,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.022769056260585785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.021893257275223732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.020283309742808342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.01951954886317253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0150186438113451,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01434130035340786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.012997607700526714,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009044409729540348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2257997989654541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.20301946997642517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19423991441726685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.17116889357566833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10530448704957962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09642394632101059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12401852011680603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.1123625859618187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10746617615222931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08944348245859146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.08399651944637299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0638689175248146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054264314472675323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05109007656574249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05034219101071358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032092299312353134,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.0271917674690485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026480693370103836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02412840910255909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02363264374434948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017424728721380234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.018006546422839165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016407500952482224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013257740996778011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15382196009159088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14381498098373413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1396910846233368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12702886760234833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07270950824022293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06881019473075867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08308828622102737,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07620883733034134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07357731461524963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06495369225740433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.062314413487911224,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.042544521391391754,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03672734647989273,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.035143643617630005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03476990759372711,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021390043199062347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018609371036291122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018298383802175522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017106514424085617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.016874471679329872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011657150462269783,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012097058817744255,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011129972524940968,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008852720260620117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23629136383533478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.221998929977417,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2167816162109375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1970624476671219,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11181584000587463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10657082498073578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12595754861831665,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11568234115839005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.1129145696759224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10006839781999588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0960417091846466,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06434372812509537,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05548098310828209,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05371786653995514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05330934748053551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.032316654920578,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.027896175161004066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027507085353136063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025547264143824577,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025291267782449722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017312465235590935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017253439873456955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016725366935133934,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011848049238324165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.17.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.23277296125888824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20573410391807556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1946689784526825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.17207078635692596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10690053552389145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09603912383317947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12889841198921204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11631854623556137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.11022789776325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.09005008637905121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.085513174533844,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06579466909170151,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05594460293650627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05167718231678009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.050617765635252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.033234138041734695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.027429264038801193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.026894068345427513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.024185629561543465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.023510124534368515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0182279571890831,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.018219564110040665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.0167915690690279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013175844214856625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11159697920084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10284176468849182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09691385924816132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08699440956115723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05219624936580658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04757829010486603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06522989273071289,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05899149924516678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05303889140486717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04585108160972595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04416908323764801,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.033286046236753464,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028333332389593124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025280551984906197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02452143095433712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016733825206756592,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013424481265246868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012912550009787083,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012190655805170536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011723286472260952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.008938467130064964,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009078599512577057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00782180018723011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006471633445471525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09080873429775238,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0825132355093956,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.074677474796772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06732600182294846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04180391505360603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03631275147199631,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.05631629005074501,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0514843687415123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042789045721292496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03666194528341293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03579147905111313,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02847900614142418,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.024534741416573524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02019139565527439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019041620194911957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014246593229472637,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010592938400804996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.00984676368534565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009584269486367702,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008834848180413246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007452117744833231,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007347347680479288,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005969054065644741,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004772136453539133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.21817098557949066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19805368781089783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18795980513095856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17003194987773895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10196997970342636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09285135567188263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12577274441719055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1121627613902092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10425809025764465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08868373185396194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08565255254507065,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0645514577627182,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053657468408346176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.048984818160533905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04783867299556732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03234352543950081,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02512260526418686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024219023063778877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.022392522543668747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021634038537740707,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016693469136953354,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015652792528271675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01456428226083517,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009827886708080769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19611601531505585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17451535165309906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16691315174102783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14427968859672546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0910203754901886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08288411796092987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10581089556217194,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09621437638998032,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09244778007268906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07538393139839172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07042215764522552,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05457882955670357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04667678102850914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04434116557240486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04377315938472748,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027627406641840935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02394750341773033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.023392587900161743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02099853940308094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020642446354031563,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015439669601619244,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016207506880164146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01468756515532732,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012435068376362324,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14265544712543488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13327494263648987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12921646237373352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.11749128997325897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06741973012685776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06365444511175156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07725055515766144,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07099489867687225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06826197355985641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.060155946761369705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05766630917787552,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03952312469482422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0341607928276062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.032549675554037094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03216953203082085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019820544868707657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017160383984446526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01685040071606636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015736378729343414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015500363893806934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010679199360311031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011069446802139282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01013621874153614,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007981355302035809,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22432978451251984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21059300005435944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20538759231567383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.186675027012825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10603143274784088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10088574141263962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11982519179582596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.1101493239402771,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10717980563640594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09490736573934555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09087194502353668,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061201393604278564,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05278884619474411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05094105750322342,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.050502434372901917,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030659247189760208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026339607313275337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02594776079058647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02409440465271473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.023816602304577827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016307314857840538,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01617657206952572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015671780332922935,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010914131067693233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.18.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22135470807552338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19446787238121033,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1830441653728485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16147533059120178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10138344764709473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09030714631080627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12372703105211258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11153721064329147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10473138839006424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08512913435697556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08101708441972733,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06333307176828384,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05381900072097778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0491536520421505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04799271002411842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03215419501066208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026349464431405067,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025791587308049202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023273451253771782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022537147626280785,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017878437414765358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01791355572640896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.016312770545482635,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013245079666376114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11465656012296677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10600828379392624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09989749640226364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0902431458234787,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053645115345716476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0489417165517807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06711427122354507,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.060852937400341034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05449691414833069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04747826233506203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.045961130410432816,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03420788794755936,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.029157904908061028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025898372754454613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.025087988004088402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017144372686743736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013635821640491486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013091490603983402,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012428264133632183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011927100829780102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00912641640752554,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009099568240344524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007961972616612911,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006304722744971514,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.08965000510215759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08232685178518295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07426565140485764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06724289059638977,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04138024151325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0359480194747448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.056460265070199966,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.051676955074071884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.042223282158374786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0367119237780571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03599132224917412,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02862909622490406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02458825148642063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02000809647142887,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018770568072795868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014324580319225788,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010489951819181442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009705272503197193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009572271257638931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008795415051281452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007511377800256014,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.0073248897679150105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005949682090431452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004749493673443794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22832971811294556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2091153860092163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.20005400478839874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.18111038208007812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1068718433380127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09854882955551147,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12875410914421082,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11564058065414429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10896963626146317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09360790997743607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08994371443986893,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06586138904094696,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05534180998802185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123921111226082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.050247516483068466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.032931022346019745,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026108600199222565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025332389399409294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023362506181001663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022718077525496483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01687576062977314,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015892764553427696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015032531693577766,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00970155093818903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21137022972106934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1925043761730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.18610072135925293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1632358729839325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09859667718410492,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09163976460695267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11444369703531265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10292671620845795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09996694326400757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08387401700019836,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0796528160572052,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.059008631855249405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05010414123535156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04815682768821716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.047679923474788666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02986222133040428,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.026172997429966927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025721712037920952,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02343345619738102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.0231548510491848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016754725947976112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01785171777009964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016134297475218773,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01397703867405653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.13911974430084229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.130160853266716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.12640529870986938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.11501024663448334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06569506973028183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06214374676346779,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07496024668216705,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.06896717846393585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06651759892702103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.05869591608643532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05628379061818123,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03827868029475212,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.033143602311611176,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03168552368879318,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03133014962077141,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019200792536139488,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.016632873564958572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01634667068719864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015251516364514828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015037299133837223,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010289926081895828,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010616136714816093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.009803997352719307,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007561687845736742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.21969464421272278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.20638231933116913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.20135000348091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18314434587955475,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1038883775472641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09891800582408905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.11729646474123001,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10784762352705002,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10504719614982605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09307779371738434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.089162178337574,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.05984938517212868,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05170333757996559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.04993525519967079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.04951624944806099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030045464634895325,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02594025991857052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025562485679984093,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02376994490623474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02350855991244316,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016147594898939133,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01608770340681076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015559244900941849,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011065609753131866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.19.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20882654190063477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18293501436710358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1708936095237732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1502760797739029,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09569812566041946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08449085056781769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11814698576927185,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10681642591953278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09894713014364243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07988262176513672,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0761018916964531,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06059737503528595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05178295448422432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.046530455350875854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04524950683116913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03089858964085579,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025215450674295425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02453891932964325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02227173186838627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02141343243420124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017463907599449158,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017549382522702217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015780964866280556,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.013132983818650246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11444326490163803,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1055014580488205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09825535863637924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0891423374414444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053389087319374084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04801574721932411,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06906714290380478,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06236892566084862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05436888337135315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0473749116063118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04616570472717285,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03520072624087334,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02987632527947426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.0258669164031744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024836190044879913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01766885630786419,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01373202446848154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013081732206046581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012574264779686928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011948813684284687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009422007016837597,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009440392255783081,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007976274937391281,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006634380668401718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09047260880470276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08232685923576355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07258374243974686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06598429381847382,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04149811714887619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.035029057413339615,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0590379424393177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053789906203746796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04254136607050896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0367581807076931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.036301515996456146,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.029818270355463028,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.025577928870916367,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.020057296380400658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.018573004752397537,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014916702173650265,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010609252378344536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009675871580839157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009685897268354893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008738551288843155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00780309597030282,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00764507194980979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005905516445636749,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004975953139364719,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22678525745868683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2049216777086258,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19451342523097992,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1758013367652893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10528506338596344,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09566793590784073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12972676753997803,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11574698984622955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10772284865379333,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09136522561311722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08834898471832275,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06667500734329224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05533298850059509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05051372945308685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049303170293569565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03336772322654724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025823956355452538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.024896109476685524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02298101596534252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02219715341925621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.0171296838670969,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016026947647333145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014846895821392536,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00993427261710167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.22984875738620758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.2021685689687729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19181154668331146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16082192957401276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10723299533128738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0965796411037445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12582704424858093,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11445076018571854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10925163328647614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08504299819469452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.079674132168293,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06471829116344452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05526352301239967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.051943499594926834,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05112643912434578,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.032539453357458115,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.027630910277366638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02689254656434059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.023370549082756042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.022835325449705124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.017570000141859055,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.018266256898641586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.016515854746103287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.013427401892840862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1456148475408554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1367599219083786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13323040306568146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12116726487874985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06882210820913315,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06545256078243256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07771433144807816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07165767252445221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06958616524934769,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06164563074707985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.05905027315020561,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03964614123106003,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03439237177371979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.033128201961517334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.032827407121658325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.019848518073558807,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017282966524362564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.017022754997015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015853779390454292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015668600797653198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010559497401118279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010853266343474388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01013503223657608,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007590027526021004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22867681086063385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.21522530913352966,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2104169726371765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19131822884082794,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10803673416376114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10315576940774918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12112215906381607,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11150986701250076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10908965766429901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09691968560218811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09275484830141068,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061633262783288956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05334612727165222,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05178083851933479,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05141318216919899,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.030911065638065338,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.026616882532835007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.026269732043147087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024352600798010826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02411971427500248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01639235019683838,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016056003049016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015872187912464142,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010563229210674763,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.20.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20761169493198395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18257753551006317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17165516316890717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15128041803836823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09508275985717773,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08455172181129456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11502408981323242,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10470683872699738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09812071174383163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07946012169122696,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07559830695390701,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05888964235782623,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05035287141799927,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04591505974531174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044816240668296814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029812291264533997,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024339772760868073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02380193956196308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021334176883101463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020627474412322044,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016428286209702492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016223173588514328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014982105232775211,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011625177226960659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10928948223590851,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10006524622440338,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09303509443998337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08444321900606155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.050899043679237366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04556712508201599,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.0658453032374382,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05945511534810066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05192425101995468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.044911228120326996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0437929630279541,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03360670059919357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.0284200981259346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024586215615272522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023609735071659088,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016865352168679237,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012933803722262383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012287275865674019,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011757158674299717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011147325858473778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009004323743283749,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008747897110879421,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007616496179252863,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005942426156252623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09122080355882645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0827295333147049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07359453290700912,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06687337160110474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04202055186033249,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03561447188258171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.058935899287462234,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.053484562784433365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04306158050894737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03705049678683281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03659228980541229,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.02994987927377224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0255876611918211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02033252641558647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.01890396699309349,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.014943400397896767,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.010692759416997433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.009786752983927727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.009723183698952198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.008815012872219086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.007780096028000116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00760724488645792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.005941588431596756,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.004923895932734013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.21679089963436127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19406531751155853,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18209117650985718,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16445887088775635,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1000591441988945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08938778936862946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12616044282913208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11263936012983322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.1027204692363739,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08627913892269135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08350113779306412,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06491731852293015,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053830377757549286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04797002300620079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04648280143737793,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03252711519598961,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.024591311812400818,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02347838319838047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.021786458790302277,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.020803524181246758,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01671263948082924,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015561647713184357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014025588519871235,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00958433747291565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2303474247455597,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.200810506939888,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.19055958092212677,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.16070528328418732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10687829554080963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09497074782848358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.12309370189905167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.11327061057090759,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10911892354488373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.08520635962486267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07752732932567596,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06319588422775269,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.054441068321466446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.05147664248943329,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.05077638849616051,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03168822452425957,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.026984339579939842,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.026227418333292007,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02272414043545723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.022253308445215225,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.01691633090376854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.017261743545532227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01593218930065632,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012141730636358261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14371661841869354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.13515472412109375,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13193932175636292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12005634605884552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.06801272928714752,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0648375153541565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07643576711416245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07048289477825165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.06867460906505585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.060972027480602264,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.058374855667352676,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.03906713053584099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03381593897938728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03271247446537018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.032450467348098755,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.01957164704799652,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.017039747908711433,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.016797924414277077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015634218230843544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.015469217672944069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010480672121047974,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010633102618157864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01012298185378313,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007406961638480425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23519429564476013,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22171618044376373,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.21705223619937897,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.19742456078529358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1113191619515419,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10647161304950714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12421892583370209,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11450593173503876,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11228305846452713,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09994170069694519,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0955977588891983,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06342466175556183,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05481244996190071,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.053373727947473526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05303028225898743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03174586966633797,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0274603720754385,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02712080255150795,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025146547704935074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024931227788329124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01685454323887825,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016577906906604767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01638275384902954,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010981686413288116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.21.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2052726000547409,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18048812448978424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.16935867071151733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.15033024549484253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0936426892876625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08314555883407593,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11404795944690704,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10393796861171722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09687253832817078,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07892251014709473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07514623552560806,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.057932473719120026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.049950361251831055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04526190087199211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.044099144637584686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029377024620771408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02403027005493641,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023432869464159012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021204207092523575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020439133048057556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016207829117774963,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016114214435219765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014700859785079956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011512459255754948,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.10957325994968414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09994599968194962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09271310269832611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08412954956293106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.050933849066495895,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.045396242290735245,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06604039669036865,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.05968862771987915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.052117422223091125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.044849272817373276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04372585937380791,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03375091031193733,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02858457900583744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.024624032899737358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023603098466992378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01696127839386463,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.012966522015631199,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012304143980145454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01177222654223442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011138373054564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009057406336069107,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008805310353636742,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007635244634002447,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005993344821035862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09877453744411469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08956434577703476,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07973670959472656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0725289136171341,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.045579712837934494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03869170323014259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06360765546560287,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05781104788184166,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04672587662935257,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.040120042860507965,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03957948833703995,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.032331496477127075,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027652788907289505,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02208142727613449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020574579015374184,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016180474311113358,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011671765707433224,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010717642493546009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010614946484565735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009668298065662384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008505980484187603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00833604484796524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006594098638743162,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.0055103180930018425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2146129459142685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.19117869436740875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1782289296388626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16099302470684052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.09880749136209488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.08744488656520844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.12621107697486877,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11253862082958221,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10159089416265488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08476818352937698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08218155801296234,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06464135646820068,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.053708694875240326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.04738851636648178,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04576430097222328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.032327935099601746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.024311577901244164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.023111021146178246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02147120237350464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02039506658911705,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01654907315969467,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01550542563199997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.013749765232205391,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009554313495755196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21571846306324005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18611060082912445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17531917989253998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14945052564144135,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.1008707731962204,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09003767371177673,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11816170811653137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10780710726976395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.1029682606458664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07973866164684296,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0729612410068512,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06060759350657463,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05197511985898018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04882344603538513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04804620519280434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.030374085530638695,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02580353058874607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.025104764848947525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.021735278889536858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02121789939701557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016415836289525032,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016898665577173233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015385137870907784,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0122162364423275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15724799036979675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14801055192947388,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14456278085708618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13154108822345734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07440970838069916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07095904648303986,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08344663679599762,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07703141123056412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07512904703617096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06670434027910233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06390923261642456,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.0426163449883461,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.036944154649972916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.035757336765527725,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035480011254549026,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021316412836313248,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01857135072350502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018309658393263817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017029915004968643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01685103215277195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011337440460920334,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01150318793952465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010952526703476906,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007916001603007317,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.23904350399971008,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2253643125295639,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2206437885761261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20078004896640778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11306136846542358,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10818864405155182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12597690522670746,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11627401411533356,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11410269141197205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1015218049287796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09717012196779251,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06427565962076187,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05563081428408623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05418301373720169,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05384523794054985,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0321015864610672,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0277982447296381,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02745252102613449,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025441676378250122,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02522149682044983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.016887299716472626,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01666787452995777,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016404885798692703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010895858518779278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.22.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21165591478347778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.18738839030265808,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17656877636909485,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1568971425294876,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.0969102680683136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.0867493748664856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.11748203635215759,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.1068744957447052,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09991614520549774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08211133629083633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07816799730062485,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05971162021160126,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05125107616186142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0467614084482193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04563642665743828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.0302292350679636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024670062586665154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02412590943276882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021808689460158348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021095577627420425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016624247655272484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01631348580121994,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015167895704507828,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011537257581949234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11144105345010757,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10212317854166031,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09526549279689789,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.0864814817905426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05190658941864967,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04660457372665405,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06652797758579254,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06018252298235893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.0529869981110096,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04579433798789978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.044573768973350525,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.033835429698228836,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028793832287192345,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025053590536117554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02409978397190571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.016952916979789734,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013133362866938114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012500988319516182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011926903389394283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011330543085932732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00897444412112236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008796442300081253,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007630597334355116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.005924706347286701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09834861755371094,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08901286870241165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08058246970176697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07312973588705063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04527607560157776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03916103392839432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06106571853160858,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.0556943379342556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04643521085381508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.039720963686704636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03891979157924652,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03095146454870701,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02656635455787182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02190260961651802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020674949511885643,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.015520544722676277,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011497283354401588,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010694186203181744,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010401714593172073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.00962082203477621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008115554228425026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00796891562640667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006500585936009884,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005228078458458185,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22936303913593292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20820878446102142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19763407111167908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17858092486858368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.106652170419693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0970284715294838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1297520250082016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11741314083337784,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10894081741571426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09268859773874283,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0893661305308342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06626816838979721,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.056016065180301666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05111228674650192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04990869015455246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033119283616542816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026141654700040817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025206558406352997,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023295940831303596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02249949984252453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.016994085162878036,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016197707504034042,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014908474870026112,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01002773828804493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.2108493149280548,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18377168476581573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.17388375103473663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14398780465126038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0971338301897049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08699032664299011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11393299698829651,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10438046604394913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09974299371242523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07697572559118271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07006049156188965,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05831246078014374,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.05016935244202614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04685966670513153,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04606541618704796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.029204130172729492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.024611441418528557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.023848094046115875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02069205790758133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.02015066333115101,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015600213780999184,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.015930181369185448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.014527887105941772,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011241447180509567,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16388827562332153,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1541646271944046,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15059228241443634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13702112436294556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07755082100629807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0739508718252182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08709289133548737,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08020004630088806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07829032838344574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06949421763420105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.0665753185749054,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04442601650953293,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.038472916930913925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037278834730386734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03699669614434242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02231348305940628,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019385285675525665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01911500096321106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.0177652295678854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017593974247574806,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012034471146762371,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012025880627334118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01165598165243864,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008325144648551941,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24376733601093292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22961558401584625,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2248132824897766,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20449243485927582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11534594744443893,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11028263717889786,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12855149805545807,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11855839192867279,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11639999598264694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10343199223279953,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09889744967222214,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06556588411331177,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05673935264348984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05528217926621437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05493824928998947,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03280350938439369,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028353199362754822,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.027995241805911064,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.025915885344147682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025698309764266014,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017327969893813133,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.016974076628684998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01683942787349224,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011072498746216297,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.23.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.21763044595718384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1923941969871521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18110594153404236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16124969720840454,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09962042421102524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08897606283426285,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.120120570063591,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11001227796077728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10273993015289307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.0843755230307579,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08043208718299866,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.061254099011421204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05274949595332146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.0480387881398201,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04686591029167175,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030863473191857338,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025283673778176308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02469497174024582,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022322557866573334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021569600328803062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01666448265314102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01665089838206768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015107650309801102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.011665080673992634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11982377618551254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1102212518453598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10303369909524918,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09354404360055923,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05602928623557091,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05050589516758919,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07108020782470703,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06468871235847473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05703223869204521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.049530621618032455,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.048145826905965805,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.036255158483982086,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030983267351984978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02705051563680172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.026059577241539955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018155114725232124,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014207431115210056,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013546726666390896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012927585281431675,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012312479317188263,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009571581147611141,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009526669979095459,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008186751045286655,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006474317982792854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.105857253074646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.09681237488985062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08756309002637863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07963361591100693,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.049108684062957764,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.0425257682800293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0668167695403099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.060869138687849045,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.050196968019008636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04343812167644501,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04267967864871025,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03392903879284859,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.029076386243104935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.0237272996455431,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022276148200035095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016949808225035667,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012385053560137749,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011477666907012463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011267125606536865,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010363537818193436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008854063227772713,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008587241172790527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007035992108285427,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005521143786609173,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23770567774772644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21765241026878357,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.2083035111427307,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1884317398071289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11104048788547516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.1023450717329979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13261345028877258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12016896903514862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11312233656644821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09735787659883499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09344839304685593,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06752979010343552,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05735821649432182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053194575011730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05217171087861061,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03375744819641113,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027144575491547585,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02631070837378502,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.02430320531129837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023617621511220932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017297660931944847,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016562573611736298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015533704310655594,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01016708742827177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1822643131017685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1650846302509308,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15686260163784027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.1310717761516571,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0854109600186348,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.0780898779630661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10173016786575317,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09326981008052826,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08658351749181747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07047121971845627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0642717108130455,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05195440351963043,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04468100890517235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04109637439250946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04020283371210098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.025976350530982018,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.021295061334967613,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.020598160102963448,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018290936946868896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.017697649076581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013676178641617298,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.013509802520275116,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012464023195207119,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00896008126437664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1644883006811142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1546863466501236,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15115080773830414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.13738568127155304,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07785758376121521,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07422785460948944,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08722874522209167,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08050209283828735,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07861840724945068,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06969896703958511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06669887900352478,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.044481806457042694,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03859532251954079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03740832582116127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.037124037742614746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02226800099015236,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01938902959227562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01912275142967701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.017748164013028145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.017569800838828087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011835074983537197,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01194112841039896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01145770400762558,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008156314492225647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.24699078500270844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23270951211452484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2278493493795395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20718933641910553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11691351979970932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.1118030995130539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13060861825942993,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12017872929573059,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11802347749471664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10484223067760468,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.1002972275018692,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06657490879297256,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.057536207139492035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05607863888144493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.055738404393196106,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033394187688827515,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028808271512389183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02845500037074089,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026330342516303062,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026110786944627762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01778838224709034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01731196418404579,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01731100305914879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011376766487956047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.24.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.2215447723865509,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19618546962738037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1850007325410843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16475047171115875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.1013823002576828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09081171452999115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12227802723646164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11167708784341812,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10460495948791504,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08610530942678452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0819515660405159,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06198311969637871,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05348736792802811,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.04883664473891258,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04767434298992157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.031283020973205566,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.025610540062189102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02503371611237526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.022622372955083847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02187088504433632,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.016980202868580818,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016727423295378685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.01545341033488512,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0115788159891963,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12556305527687073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.1159956231713295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10893036425113678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09871136397123337,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0588366873562336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05336252599954605,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07406643033027649,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06730064749717712,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.059843894094228745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.052128471434116364,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05057583749294281,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03771689906716347,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.03223186358809471,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02836601994931698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.027398614212870598,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018879787996411324,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014800465665757656,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01413826085627079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013466855511069298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012852605432271957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00989446323364973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.00976674072444439,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00851232185959816,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006496401038020849,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10687704384326935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.09802205115556717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08918632566928864,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.08103649318218231,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.049645498394966125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.043320197612047195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.0663713738322258,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.060791224241256714,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05062929913401604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04395134374499321,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04291936010122299,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03364581987261772,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02906634286046028,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02394874580204487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.022595783695578575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016815830022096634,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012461531907320023,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011590085923671722,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.01133390050381422,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010473296977579594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008740531280636787,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00853308942168951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007042492739856243,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.00542002497240901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.239474356174469,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.21975542604923248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.2106943130493164,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1905088871717453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11187077313661575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10350967943668365,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1340796798467636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12072005122900009,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11383438855409622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0981183797121048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09425124526023865,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06833413988351822,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.0576237253844738,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.053602997213602066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.052614402025938034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.034064799547195435,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02730974368751049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02648700773715973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.024464676156640053,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.023826181888580322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01750899851322174,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016577132046222687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01571711339056492,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010110314004123211,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.19715285301208496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17291495203971863,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1650565266609192,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.136726513504982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09188529849052429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08323013037443161,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10596141219139099,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09645688533782959,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09369111061096191,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07316645234823227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06495869904756546,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05389346927404404,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04622245207428932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04414921998977661,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04363996908068657,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.027014337480068207,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.022814135998487473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02226768620312214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01903679035604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018694404512643814,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014341474510729313,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014087200164794922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013615787029266357,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.009435413405299187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.16994985938072205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15970930457115173,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.15601615607738495,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14184722304344177,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0805244967341423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07675040513277054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0902675986289978,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08324528485536575,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08130747824907303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07200834900140762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06889088451862335,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.046129852533340454,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03994375467300415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03870568424463272,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03841806575655937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.023116545751690865,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020098570734262466,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019830169156193733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018395040184259415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01821562834084034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01236339844763279,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012440857477486134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011966140940785408,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008563359268009663,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.25007277727127075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23540708422660828,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2304924726486206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.209419846534729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11847733706235886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11326760798692703,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13172279298305511,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12175863236188889,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11957496404647827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10608039051294327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.10131092369556427,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06711410731077194,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05823390185832977,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.05675370618700981,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.056398868560791016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03351031243801117,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.029027877375483513,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02866833098232746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026475690305233,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026250038295984268,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017450524494051933,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017264489084482193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016953660175204277,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011098220013082027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.25.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22341962158679962,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.19779229164123535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18608558177947998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16591614484786987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10212691873311996,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09126339107751846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12343217432498932,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11322055757045746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10547378659248352,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08701349794864655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08286767452955246,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06288120150566101,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.054267920553684235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.049180008471012115,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04791262000799179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03156855329871178,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02575099654495716,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025119254365563393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02278914488852024,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.021966073662042618,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.0168871209025383,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01683790795505047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015216715633869171,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.0115489661693573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12053371965885162,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.11076760292053223,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10318152606487274,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.093513622879982,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.056388288736343384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.05059025064110756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07188374549150467,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06559693813323975,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05744076520204544,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04981626942753792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.048390377312898636,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03664573282003403,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.031455881893634796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.027232738211750984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.026162054389715195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.018347304314374924,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014282265678048134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013571036979556084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012987343594431877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012313530780375004,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00958429928869009,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009602971374988556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008095034398138523,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.00645834906026721,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10353386402130127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0944262221455574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08451580256223679,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07670380175113678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04793551564216614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.04102899506688118,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06642840802669525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.06052868440747261,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04905562102794647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.042328137904405594,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.041505519300699234,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0335739329457283,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02885543182492256,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02319573424756527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.021646760404109955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016795940697193146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012149459682404995,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011178635992109776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011039599776268005,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010068141855299473,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008812550455331802,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008530905470252037,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006894449237734079,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005481294821947813,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23001907765865326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20809012651443481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.1968235969543457,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17770859599113464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.1067451760172844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09663660824298859,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13227549195289612,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11854508519172668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10938778519630432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09268905967473984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08947689831256866,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06777966022491455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05658711493015289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05123372748494148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049881577491760254,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.033908549696207047,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026182422414422035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02514859102666378,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023289721459150314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02240009233355522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017397766932845116,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016305483877658844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014977026730775833,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009975658729672432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18919837474822998,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.17131733894348145,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.16575421392917633,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14084136486053467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08916810154914856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08256521075963974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10014063864946365,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09214363992214203,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09038402140140533,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07386420667171478,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06727684289216995,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.051175300031900406,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.044151365756988525,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04279990494251251,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04248662292957306,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.025581825524568558,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02205626666545868,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.021664096042513847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018968136981129646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.018748996779322624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.013436826877295971,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.013422160409390926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.012991671450436115,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.008961319923400879,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.17453797161579132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.16400179266929626,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.16019724309444427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14550961554050446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.08278516680002213,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07890249043703079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09273967146873474,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08559832721948624,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.0836155116558075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0739915668964386,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.07065751403570175,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04734564572572708,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.041043464094400406,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03977876156568527,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.039477959275245667,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.023681458085775375,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020611200481653214,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.020329318940639496,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.018839513882994652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018649715930223465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012530026957392693,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012689477764070034,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01211528293788433,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008654211647808552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.2512197196483612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.23645330965518951,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.23150409758090973,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.21027915179729462,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.11918137967586517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11390683054924011,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.13286268711090088,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.12249982357025146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.12028618156909943,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.10664454847574234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.10184229910373688,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06776771694421768,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.0586634986102581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.057152822613716125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0567997507750988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.033883705735206604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.029367459937930107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.029003577306866646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.026795823127031326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.026569750159978867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01788564771413803,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01764919050037861,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01738482527434826,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011606301181018353,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.26.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22692427039146423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20162063837051392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.19002698361873627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1692453920841217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10395738482475281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09318237006664276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.12549404799938202,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11499937623739243,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10729505121707916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08877211809158325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.08449645340442657,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06395233422517776,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.0551777146756649,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05010344088077545,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04885312542319298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.03233486786484718,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026354603469371796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025724049657583237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023392533883452415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022584237158298492,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017627809196710587,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017340756952762604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015992915257811546,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012071800418198109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.12452154606580734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.11453695595264435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.10809440165758133,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.09805474430322647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.05838008224964142,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.053114622831344604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07246813923120499,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06587137281894684,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05941365286707878,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.05149584263563156,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.0498381033539772,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03697812929749489,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.031559258699417114,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.028150558471679688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.027310607954859734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01852261647582054,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014705331064760685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01410877425223589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013334398157894611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.012790704146027565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009739626199007034,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009637013077735901,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.008491975255310535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0064606983214616776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.10733161121606827,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.0989459753036499,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.09235842525959015,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0837026834487915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.05014695227146149,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.04509536176919937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06361188739538193,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05822722986340523,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.05106978118419647,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.04435816779732704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.04289370775222778,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03225512057542801,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02784596011042595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.024184204638004303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.023253321647644043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016180627048015594,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.012574711814522743,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.011957300826907158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.011417492292821407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.010832133702933788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008471757173538208,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008316163904964924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.007201797794550657,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005438519176095724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22618617117404938,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20783376693725586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19920216500759125,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.18024316430091858,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10609455406665802,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09800407290458679,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.125311478972435,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11450386047363281,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10792074352502823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09308404475450516,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08924183249473572,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06385636329650879,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05472434312105179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05085771530866623,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04990743100643158,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.031863462179899216,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.02589772827923298,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02509617619216442,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023191772401332855,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022561104968190193,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01634899154305458,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.015685463324189186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.014880356378853321,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.009533638134598732,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.20709311962127686,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18371646106243134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1756037026643753,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.14739514887332916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.09714093059301376,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.08835341036319733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.11171499639749527,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10243193060159683,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.09894679486751556,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07798339426517487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07212778180837631,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.057404011487960815,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04943528771400452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04704694449901581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.046497926115989685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.02893695794045925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.02502412721514702,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.02444585971534252,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.021348239853978157,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020963555201888084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.015841927379369736,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.016442058607935905,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.01509474590420723,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.012123816646635532,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.1696636974811554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15925085544586182,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1554262638092041,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.14101669192314148,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.0805472731590271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07662193477153778,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.09046194702386856,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0834740698337555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.08137940615415573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.07186915725469589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06864246726036072,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04620400816202164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04008869826793671,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.038752127438783646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.038441333919763565,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02313356287777424,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.02013331465423107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019838061183691025,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01839279942214489,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018199166283011436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012288039550185204,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.012490564025938511,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011849011294543743,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.00861911941319704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.243917778134346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.22943468391895294,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.22448599338531494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.20380203425884247,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.1157817617058754,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.11056102067232132,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.12916427850723267,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11916787177324295,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.11689942330121994,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.1034947857260704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09887633472681046,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.06598954647779465,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.05710281804203987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.055572204291820526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.05520958453416824,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.0330083966255188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.028569214046001434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.0281955786049366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02604973316192627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.025813542306423187,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.017437513917684555,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.017210720106959343,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.016923565417528152,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011341173201799393,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.27.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.22700971364974976,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.20089219510555267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.18851624429225922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.16792292892932892,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.10385089367628098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.09249718487262726,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1275073140859604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.11588882654905319,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.10735159367322922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.08841566741466522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.0842674970626831,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.06474127620458603,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.055522285401821136,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.05008291080594063,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04873017966747284,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.032673366367816925,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.026347137987613678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.025668693706393242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.023331694304943085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.022461935877799988,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.017774414271116257,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.017407815903425217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.015970418229699135,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012096922844648361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11572492867708206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10626456886529922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09801332652568817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08891233056783676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.054125506430864334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04799175262451172,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07131495326757431,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.0644075870513916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05519085004925728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04779224097728729,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04679208621382713,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03645377233624458,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.030863337218761444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02618774026632309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024978477507829666,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01823790743947029,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013802701607346535,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.013025574386119843,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.012560619041323662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011811941862106323,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009593253023922443,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009462372399866581,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007903127931058407,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.0064032673835754395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09731120616197586,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08851464837789536,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07734818011522293,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07040656358003616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04492749646306038,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.037374380975961685,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06498654931783676,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.059081148356199265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04605111479759216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.039760228246450424,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03935680538415909,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03285251930356026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02815963886678219,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021789396181702614,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020010769367218018,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016488082706928253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.011497395113110542,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.01039078924804926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010493731126189232,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009369997307658195,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008600966073572636,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008353759534657001,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006392995826900005,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005371517036110163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.23069420456886292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2078019231557846,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19543907046318054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17634005844593048,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10710982978343964,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09611407667398453,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.1343490481376648,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1203572005033493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10976240783929825,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.0927569717168808,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.0896335318684578,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.06881999224424362,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05757840350270271,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05147475376725197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.049937356263399124,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.034555140882730484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.026490207761526108,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02535529062151909,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023588579148054123,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.022594979032874107,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.017789918929338455,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01691047102212906,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015112156048417091,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010707507841289043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.21962280571460724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.18841341137886047,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1776753067970276,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.15232819318771362,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.10183048993349075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.09018845111131668,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.1187899112701416,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.10850027948617935,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.10426682978868484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.07932937145233154,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.07580279558897018,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.06088311970233917,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.052138980478048325,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.04907821863889694,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04832329601049423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.03049134463071823,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.025779221206903458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.024988699704408646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.02140943519771099,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.020911216735839844,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.016389766708016396,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.01662471890449524,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.015413561835885048,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.011755919083952904,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15797848999500275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1480463296175003,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14414793252944946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1307307481765747,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07497101277112961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07113786041736603,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08464109897613525,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.0781075581908226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07579217851161957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.0668020099401474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06375230103731155,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04328674077987671,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03749806433916092,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03608153015375137,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035743407905101776,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02166181430220604,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018741797655820847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.01844456046819687,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01710067130625248,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01689102128148079,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.011502068489789963,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011650791391730309,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011037035845220089,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008009687066078186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.22683678567409515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.2130274772644043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.2081146389245987,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.18886259198188782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.10766664892435074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.10252943634986877,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.1204918920993805,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.11119615286588669,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.10873465240001678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.09609802812337875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.09165717661380768,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.061596956104040146,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.053289253264665604,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0517064668238163,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.0513235367834568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.03083229809999466,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0266769677400589,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02631176821887493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.024330344051122665,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.02409081533551216,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01633727364242077,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.01625257357954979,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.015798402950167656,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.010893873870372772,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.28.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20614174008369446,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1810261756181717,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.168910950422287,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.14987972378730774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09427877515554428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08322176337242126,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1160864531993866,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10599948465824127,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09758887439966202,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07945822179317474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07577288150787354,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05936186760663986,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.051070112735033035,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.045723121613264084,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04437762871384621,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.030095098540186882,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.024472499266266823,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.023798203095793724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.021657794713974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.020795727148652077,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01661045290529728,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.01673768274486065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014861950650811195,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012166278436779976,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.11367612332105637,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.10432964563369751,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09678502380847931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08786752074956894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.053042247891426086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04731567203998566,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.07015542685985565,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06234882026910782,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05415944382548332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04688425362110138,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04583379998803139,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.035805508494377136,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.029868802055716515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.025623805820941925,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.024538956582546234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.01795237883925438,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013443049974739552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.012744519859552383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01222216710448265,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011546935886144638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009424896910786629,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009090684354305267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007751536555588245,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006098092067986727,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.0973631888628006,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08995675295591354,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.08168809115886688,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.07412026077508926,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04534174129366875,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.039626553654670715,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06180315464735031,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05597155541181564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04617545008659363,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.040325410664081573,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.039512503892183304,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03140662983059883,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.02674219384789467,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021911388263106346,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.020605646073818207,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.01570458523929119,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01145564578473568,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010617943480610847,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010464434511959553,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009630531072616577,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00819715578109026,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.007936128415167332,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006453374866396189,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005081410985440016,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.24874486029148102,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.2292284071445465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.22003290057182312,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.1993403136730194,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.11686625331640244,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.10830976068973541,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13931411504745483,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.1263684183359146,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11889489740133286,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.10296541452407837,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09905707836151123,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.07132161408662796,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.06042494252324104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05609932541847229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.055037982761859894,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.035800572484731674,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.028698571026325226,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.027871983125805855,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.025843419134616852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.025148339569568634,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01840903051197529,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01761748641729355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.016524985432624817,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.011030805297195911,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.18839126825332642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.16278888285160065,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.15342433750629425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.13441592454910278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.08790618926286697,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07792135328054428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.10402284562587738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.09350671619176865,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.08984015882015228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06998612731695175,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.06666959822177887,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.05339683219790459,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.04496820643544197,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.042426321655511856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.04180406033992767,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.026778001338243484,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.022253278642892838,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.021577540785074234,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.018739059567451477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.01832115463912487,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.014231531880795956,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.014290078543126583,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.013338725082576275,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.01006705779582262,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15341390669345856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14361542463302612,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13970118761062622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1266121119260788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07276876270771027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06895560771226883,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08237628638744354,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07598202675580978,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07361187040805817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06476320326328278,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.061901893466711044,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04211081936955452,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03650331497192383,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03506097197532654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03471751883625984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.021085144951939583,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.018305864185094833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.018002627417445183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.016710253432393074,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.016491355374455452,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.01124162133783102,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.011525582522153854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01075770240277052,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.008063158951699734,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.20209452509880066,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.1895570456981659,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.184932142496109,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1678933948278427,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.09669799357652664,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.09197380393743515,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.10844431817531586,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.10006734728813171,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.09762997180223465,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.08626969903707504,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.08249043673276901,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.056075092405080795,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.04892962425947189,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.047404706478118896,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.047041989862918854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.028243333101272583,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.0261384230107069,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.025827614590525627,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.02423454262316227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.024021128192543983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.015970636159181595,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.018166454508900642,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.01549572590738535,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.014622834511101246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.29.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.20571546256542206,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.1815090775489807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.17045757174491882,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.1500616818666458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.09432607889175415,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.08405863493680954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.1150156781077385,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.10463142395019531,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.09726862609386444,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.07922500371932983,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.07495211809873581,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05831105634570122,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.05018393322825432,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.045558180660009384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.04440474137663841,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.029313212260603905,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.02412106655538082,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02354668453335762,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.02121189422905445,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.02046799287199974,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.01579117961227894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.016077060252428055,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.014262380078434944,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.01144026592373848,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1095728725194931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.100243479013443,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.09128393232822418,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.08260037004947662,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.051102083176374435,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04463138431310654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.06921995431184769,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.06240910664200783,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05219741538167,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04499661177396774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04423733800649643,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.0353984534740448,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.02993728592991829,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02478559873998165,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.023427501320838928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017751535400748253,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.013142452575266361,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.01227540336549282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011954843997955322,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.011102452874183655,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.00936879962682724,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.009230590425431728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.0075102150440216064,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006275800988078117,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09547897428274155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08686858415603638,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07536287605762482,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.0684126615524292,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04400666058063507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.03634760528802872,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06484340876340866,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.058620940893888474,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.04517835006117821,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.03889043256640434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.03865814954042435,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.032974984496831894,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027953792363405228,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.021396392956376076,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019549990072846413,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016563745215535164,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01135834027081728,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.01020627561956644,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010375534184277058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009200098924338818,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.008627797476947308,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.008383037522435188,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006264181341975927,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005445764400064945,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.22474946081638336,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20137616991996765,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.18781088292598724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.16936330497264862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10393724590539932,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09239887446165085,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.13313470780849457,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.11901592463254929,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.10685907304286957,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.08965485543012619,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.08724135905504227,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.0685306265950203,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.05686498433351517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.05003635585308075,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.04826974868774414,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03444301709532738,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.025873634964227676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.02458743378520012,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.022994350641965866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.021843496710062027,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.01781383715569973,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.016799015924334526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.01471610739827156,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.010716233402490616,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.16379593312740326,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.14769664406776428,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.1425313651561737,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.12055926024913788,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.07679182291030884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.07108142971992493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.08731945604085922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.07948681712150574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.07775185257196426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.06287302076816559,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.05801357328891754,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.04490472376346588,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.03835875540971756,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.03710135444998741,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.036819204688072205,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.0225103497505188,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01961367391049862,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.019267059862613678,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.01692737452685833,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.016717446967959404,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.012195341289043427,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.012644082307815552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.011801408603787422,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.009256135672330856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.15331622958183289,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.14363162219524384,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.1398492157459259,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12690052390098572,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07327380031347275,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.0694596990942955,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.08264171332120895,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07623498886823654,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07402262091636658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06524883955717087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06245270371437073,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04276455193758011,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.03727343678474426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.03593238443136215,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.035613540560007095,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02160659246146679,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.019766774028539658,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.019496247172355652,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01829102821648121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.018106089904904366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012209240347146988,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.013711347244679928,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.011786255054175854,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.01097308099269867,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.161906898021698,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.15187588334083557,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.14817042648792267,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.1344291865825653,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07737308740615845,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.07358581572771072,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.0868956670165062,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.08015749603509903,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07815074175596237,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06899557262659073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06600962579250336,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.04486967250704765,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.039062194526195526,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.037813249975442886,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03752044960856438,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.022616958245635033,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.020649300888180733,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.02039312571287155,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.01909191906452179,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01891881786286831,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.012667782604694366,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.014124766923487186,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.012267973273992538,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.011174408718943596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.30.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.17595866322517395,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.15610191226005554,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.1475788950920105,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.12861225008964539,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.08129768073558807,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.07320835441350937,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.099279023706913,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.0883098840713501,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.08329994976520538,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.06792747229337692,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.06408126652240753,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.05014503374695778,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.042950693517923355,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.03983841836452484,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.03902662917971611,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.025673769414424896,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.021892568096518517,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.02151186391711235,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.019457319751381874,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.01897076703608036,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.014794301241636276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.015404149889945984,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.013819348998367786,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.012092744931578636,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.self_attn.q_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1061895564198494,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.0968628078699112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.0882173702120781,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07984215021133423,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.04949919879436493,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.043128401041030884,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.066900834441185,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.060333918780088425,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.05060521513223648,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.04350423440337181,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.04272889345884323,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.03416188061237335,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.028950683772563934,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.02398574724793434,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.022681990638375282,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.017103267833590508,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.01269147265702486,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.011865245178341866,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.011522394604980946,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.010716697201132774,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009003915823996067,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.008881181478500366,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.007230670191347599,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.006025675218552351,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.self_attn.k_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.09517841041088104,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.08642280846834183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.07565269619226456,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.06861930340528488,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.04396772384643555,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.036606211215257645,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.06458213925361633,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.05770289897918701,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.045150693506002426,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.038752928376197815,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.038485389202833176,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.03274136036634445,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.027539081871509552,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.02137724496424198,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.019634725525975227,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.016404755413532257,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.01133318804204464,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.010291951708495617,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.010328191332519054,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.009264237247407436,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.00861305184662342,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.00828433409333229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.006385622546076775,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.005472538061439991,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.self_attn.v_proj",
+ "numel": 4194304,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.18896484375,
+ "total_bits": 9181184.0,
+ "err": 0.2369954138994217,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.37646484375,
+ "total_bits": 9967616.0,
+ "err": 0.20976445078849792,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.62646484375,
+ "total_bits": 11016192.0,
+ "err": 0.19394990801811218,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.72021484375,
+ "total_bits": 11409408.0,
+ "err": 0.17515923082828522,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.22021484375,
+ "total_bits": 13506560.0,
+ "err": 0.10949000716209412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.68896484375,
+ "total_bits": 15472640.0,
+ "err": 0.09581846743822098,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.0316162109375,
+ "total_bits": 12715520.0,
+ "err": 0.140801802277565,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.12646484375,
+ "total_bits": 13113344.0,
+ "err": 0.12658214569091797,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.18896484375,
+ "total_bits": 13375488.0,
+ "err": 0.11301255971193314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.53271484375,
+ "total_bits": 14817280.0,
+ "err": 0.09350759536027908,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.656982421875,
+ "total_bits": 15338496.0,
+ "err": 0.09111417084932327,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.0316162109375,
+ "total_bits": 16909824.0,
+ "err": 0.07231248170137405,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.12646484375,
+ "total_bits": 17307648.0,
+ "err": 0.06061223894357681,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.22021484375,
+ "total_bits": 17700864.0,
+ "err": 0.052722275257110596,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.31396484375,
+ "total_bits": 18094080.0,
+ "err": 0.05070376768708229,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.0316162109375,
+ "total_bits": 21104128.0,
+ "err": 0.03639683127403259,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.22021484375,
+ "total_bits": 21895168.0,
+ "err": 0.027204643934965134,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.37646484375,
+ "total_bits": 22550528.0,
+ "err": 0.025729410350322723,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.53271484375,
+ "total_bits": 23205888.0,
+ "err": 0.023973781615495682,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.72021484375,
+ "total_bits": 23992320.0,
+ "err": 0.02264346554875374,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.0316162109375,
+ "total_bits": 25298432.0,
+ "err": 0.018712390214204788,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.12646484375,
+ "total_bits": 25696256.0,
+ "err": 0.01772727072238922,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.2191162109375,
+ "total_bits": 26084864.0,
+ "err": 0.015440641902387142,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.12646484375,
+ "total_bits": 34084864.0,
+ "err": 0.01116263773292303,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.self_attn.o_proj",
+ "numel": 16777216,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1878662109375,
+ "total_bits": 36706304.0,
+ "err": 0.1016402393579483,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 39852032.0,
+ "err": 0.09097951650619507,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 44046336.0,
+ "err": 0.08753809332847595,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7191162109375,
+ "total_bits": 45619200.0,
+ "err": 0.07567872107028961,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2191162109375,
+ "total_bits": 54007808.0,
+ "err": 0.0461217425763607,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6878662109375,
+ "total_bits": 61872128.0,
+ "err": 0.04225469008088112,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 50857472.0,
+ "err": 0.054787587374448776,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 52434944.0,
+ "err": 0.04818283021450043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1878662109375,
+ "total_bits": 53483520.0,
+ "err": 0.04686923325061798,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5316162109375,
+ "total_bits": 59250688.0,
+ "err": 0.0379980094730854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.65643310546875,
+ "total_bits": 61344768.0,
+ "err": 0.03609740734100342,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 67634688.0,
+ "err": 0.02759561501443386,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 69212160.0,
+ "err": 0.024853995069861412,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.2191162109375,
+ "total_bits": 70785024.0,
+ "err": 0.023935209959745407,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.3128662109375,
+ "total_bits": 72357888.0,
+ "err": 0.02371959760785103,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 84411904.0,
+ "err": 0.014634456485509872,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.2191162109375,
+ "total_bits": 87562240.0,
+ "err": 0.014866461046040058,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.3753662109375,
+ "total_bits": 90183680.0,
+ "err": 0.014681817963719368,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.5316162109375,
+ "total_bits": 92805120.0,
+ "err": 0.013609597459435463,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.7191162109375,
+ "total_bits": 95950848.0,
+ "err": 0.013487079180777073,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 101189120.0,
+ "err": 0.009156208485364914,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 102766592.0,
+ "err": 0.011941846460103989,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218841552734375,
+ "total_bits": 104334848.0,
+ "err": 0.00888765323907137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 136321024.0,
+ "err": 0.010768837295472622,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.mlp.gate_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.14864933490753174,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.139508455991745,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.13613034784793854,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.12340245395898819,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.07056599110364914,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.06713946163654327,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.07937955111265182,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.07310084253549576,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.07123634964227676,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.06288063526153564,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.06006962060928345,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.040562666952610016,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.035055696964263916,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.0339076891541481,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.03362968564033508,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.02027907967567444,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.01753162033855915,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.017274271696805954,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.015979034826159477,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01580420695245266,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.010695081204175949,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.010762694291770458,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.010303606279194355,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.007279439829289913,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.mlp.up_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1876046316964284,
+ "total_bits": 128456703.99999999,
+ "err": 0.10389047116041183,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3751046316964284,
+ "total_bits": 139466752.0,
+ "err": 0.09726250171661377,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6251046316964284,
+ "total_bits": 154146816.0,
+ "err": 0.09485975652933121,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7188546316964284,
+ "total_bits": 159651840.0,
+ "err": 0.08593438565731049,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2188546316964284,
+ "total_bits": 189011968.0,
+ "err": 0.04926689341664314,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.6876046316964284,
+ "total_bits": 216537088.0,
+ "err": 0.04681713879108429,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031276157924107,
+ "total_bits": 177997312.0,
+ "err": 0.05551265552639961,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1251046316964284,
+ "total_bits": 183506944.0,
+ "err": 0.05106944218277931,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1876046316964284,
+ "total_bits": 187176960.0,
+ "err": 0.049770206212997437,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5313546316964284,
+ "total_bits": 207362048.0,
+ "err": 0.043839775025844574,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6563023158482144,
+ "total_bits": 214699008.0,
+ "err": 0.041905198246240616,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031276157924107,
+ "total_bits": 236717567.99999997,
+ "err": 0.028441954404115677,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.125104631696429,
+ "total_bits": 242227200.0,
+ "err": 0.024597933515906334,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.218854631696429,
+ "total_bits": 247732224.0,
+ "err": 0.023779118433594704,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.312604631696429,
+ "total_bits": 253237248.0,
+ "err": 0.02358659729361534,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031276157924107,
+ "total_bits": 295437824.0,
+ "err": 0.014284290373325348,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.218854631696429,
+ "total_bits": 306452480.0,
+ "err": 0.012519893236458302,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.375104631696429,
+ "total_bits": 315627520.0,
+ "err": 0.012341500259935856,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.531354631696429,
+ "total_bits": 324802560.0,
+ "err": 0.011453012935817242,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.718854631696429,
+ "total_bits": 335812608.0,
+ "err": 0.01133302878588438,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031276157924107,
+ "total_bits": 354158080.0,
+ "err": 0.007762065157294273,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.125104631696429,
+ "total_bits": 359667712.0,
+ "err": 0.007997272536158562,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.218776157924107,
+ "total_bits": 365168128.0,
+ "err": 0.0074990964494645596,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.125104631696429,
+ "total_bits": 477108224.0,
+ "err": 0.005783628206700087,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ },
+ {
+ "key": "model.layers.31.mlp.down_proj",
+ "numel": 58720256,
+ "options": [
+ {
+ "desc": "0.05:3b/0.95:2b 32g s4",
+ "bpw": 2.1789376395089284,
+ "total_bits": 127947775.99999999,
+ "err": 0.10926377028226852,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:3b/0.75:2b 32g s4",
+ "bpw": 2.3753662109375,
+ "total_bits": 139482112.0,
+ "err": 0.09788458794355392,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.25:4b/0.75:2b 32g s4",
+ "bpw": 2.6253662109375,
+ "total_bits": 154162176.0,
+ "err": 0.09272480010986328,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 2
+ ],
+ "bits_prop": [
+ 0.25,
+ 0.75
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.4:3b/0.5:2b 32g s4",
+ "bpw": 2.7235804966517856,
+ "total_bits": 159929344.0,
+ "err": 0.08060410618782043,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3,
+ 2
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.4,
+ 0.5
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:4b/0.9:3b 32g s4",
+ "bpw": 3.2235804966517856,
+ "total_bits": 189289472.0,
+ "err": 0.050845175981521606,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.2:6b/0.8:3b 32g s4",
+ "bpw": 3.7146519252232144,
+ "total_bits": 218125312.0,
+ "err": 0.04602187126874924,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 3
+ ],
+ "bits_prop": [
+ 0.2,
+ 0.8
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 128g s4",
+ "bpw": 3.031341552734375,
+ "total_bits": 178001152.0,
+ "err": 0.06256718933582306,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:3b 32g s4",
+ "bpw": 3.1253662109375,
+ "total_bits": 183522304.0,
+ "err": 0.05570872128009796,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 3
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:4b/0.95:3b 32g s4",
+ "bpw": 3.1789376395089284,
+ "total_bits": 186668032.0,
+ "err": 0.05201271176338196,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.95
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:4b/0.6:3b 32g s4",
+ "bpw": 3.5271519252232144,
+ "total_bits": 207115264.0,
+ "err": 0.042947325855493546,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.6:4b/0.4:3b 64g s4",
+ "bpw": 3.6608973911830356,
+ "total_bits": 214968832.0,
+ "err": 0.040726907551288605,
+ "qparams": {
+ "group_size": 64,
+ "bits": [
+ 4,
+ 3
+ ],
+ "bits_prop": [
+ 0.6,
+ 0.4
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 128g s4",
+ "bpw": 4.031341552734375,
+ "total_bits": 236721408.0,
+ "err": 0.032257549464702606,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:4b 32g s4",
+ "bpw": 4.1253662109375,
+ "total_bits": 242242560.0,
+ "err": 0.027281779795885086,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 4
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:5b/0.9:4b 32g s4",
+ "bpw": 4.223580496651786,
+ "total_bits": 248009728.0,
+ "err": 0.025067172944545746,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 5,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:4b 32g s4",
+ "bpw": 4.321794782366071,
+ "total_bits": 253776896.0,
+ "err": 0.024508720263838768,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 4
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:5b 128g s4",
+ "bpw": 5.031341552734375,
+ "total_bits": 295441664.0,
+ "err": 0.016622059047222137,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 5
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:6b/0.9:5b 32g s4",
+ "bpw": 5.223580496651786,
+ "total_bits": 306729984.0,
+ "err": 0.014052143320441246,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.05:8b/0.05:6b/0.9:5b 32g s4",
+ "bpw": 5.339651925223214,
+ "total_bits": 313545728.0,
+ "err": 0.01378138829022646,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.05,
+ 0.05,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.4:6b/0.6:5b 32g s4",
+ "bpw": 5.527151925223214,
+ "total_bits": 324555776.0,
+ "err": 0.012646831572055817,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.4,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.3:6b/0.6:5b 32g s4",
+ "bpw": 5.723580496651786,
+ "total_bits": 336090112.0,
+ "err": 0.01232102606445551,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8,
+ 6,
+ 5
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.3,
+ 0.6
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 128g s4",
+ "bpw": 6.031341552734375,
+ "total_bits": 354161920.0,
+ "err": 0.009740615263581276,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:6b 32g s4",
+ "bpw": 6.1253662109375,
+ "total_bits": 359683072.0,
+ "err": 0.010195295326411724,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 6
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "0.1:8b/0.9:6b 128g s4",
+ "bpw": 6.227770124162946,
+ "total_bits": 365696256.0,
+ "err": 0.009029214270412922,
+ "qparams": {
+ "group_size": 128,
+ "bits": [
+ 8,
+ 6
+ ],
+ "bits_prop": [
+ 0.1,
+ 0.9
+ ],
+ "scale_bits": 4
+ }
+ },
+ {
+ "desc": "1.0:8b 32g s4",
+ "bpw": 8.1253662109375,
+ "total_bits": 477123584.0,
+ "err": 0.008230828680098057,
+ "qparams": {
+ "group_size": 32,
+ "bits": [
+ 8
+ ],
+ "bits_prop": [
+ 1.0
+ ],
+ "scale_bits": 4
+ }
+ }
+ ]
+ }
+ ],
+ "last_module_idx": 66,
+ "base_perplexity": 6.128923067853044
+}
\ No newline at end of file
diff --git a/model-00001-of-00008.safetensors b/model-00001-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..86076624c99db1803a123dc5b38e55cc937f8663
--- /dev/null
+++ b/model-00001-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a326e1e4738e5542c4c7fcbd4a98042b656f710a58444aec9b65c03106a173f
+size 1889587040
diff --git a/model-00002-of-00008.safetensors b/model-00002-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3a4575b4fcbe390493f7c35a6a03c7488318e9a5
--- /dev/null
+++ b/model-00002-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd3780c719220b990e284d46d725de1e897a41d2a2c9cb91412e67eb37febd81
+size 1946243936
diff --git a/model-00003-of-00008.safetensors b/model-00003-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..26c0552acab030f8924b14599144af19127a96a0
--- /dev/null
+++ b/model-00003-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f193bbbdbf45f96e70119c361fb65d577bd96fbc9929f6f8b8ef381f06dae78d
+size 1979781432
diff --git a/model-00004-of-00008.safetensors b/model-00004-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2fa197250fa0dd5be92661146f603fe873c37fec
--- /dev/null
+++ b/model-00004-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57aa1516be6c70c2dc0bdc9358aa3fc21a0a9e1f495b4a489be0892a77fcc169
+size 1946243984
diff --git a/model-00005-of-00008.safetensors b/model-00005-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ea410bcc51cc02ad83eccd126f6af59eceddaeeb
--- /dev/null
+++ b/model-00005-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6d3de7f15f72b7efaea9fa7b4c3a8250c2b3977b44dbf6a541c9e833d87dafe
+size 1979781448
diff --git a/model-00006-of-00008.safetensors b/model-00006-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..67ee057c33802be2980c6e6716d75e1a51ff213c
--- /dev/null
+++ b/model-00006-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7802127af417a113424be223deaeb2dea0bd74518a0ae81a9bf8b7e90592f4f
+size 1946243984
diff --git a/model-00007-of-00008.safetensors b/model-00007-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a0265feba24243083285b9be1887e82620764590
--- /dev/null
+++ b/model-00007-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aa156167951408feeb432b4a459cab2a7b370b6d9bae8552e9438333ea7b73c
+size 1979781448
diff --git a/model-00008-of-00008.safetensors b/model-00008-of-00008.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ff3b27d743060732b3e3098b0428c6777d2e7ab3
--- /dev/null
+++ b/model-00008-of-00008.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc926df799ef165ad0dd8b4b648f6860ed06366583950a9922ada27633d4c477
+size 815834680
diff --git a/model.safetensors.index.json b/model.safetensors.index.json
new file mode 100644
index 0000000000000000000000000000000000000000..fbc869b880f0c7287847c72de41d71522f62b685
--- /dev/null
+++ b/model.safetensors.index.json
@@ -0,0 +1,298 @@
+{
+ "metadata": {
+ "total_size": 14483464192
+ },
+ "weight_map": {
+ "lm_head.weight": "model-00008-of-00008.safetensors",
+ "model.embed_tokens.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.input_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.mlp.down_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.mlp.up_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.input_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.mlp.down_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.mlp.up_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.10.input_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.10.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.input_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.11.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.input_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.12.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.12.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.12.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.12.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.13.input_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.13.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.input_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.14.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.input_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.15.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.input_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.16.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.17.input_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.17.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.17.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.17.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.17.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.17.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.17.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.17.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.17.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+ "model.layers.18.input_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.18.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.input_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.19.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.2.input_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.mlp.down_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.mlp.up_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.20.input_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.20.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.input_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.21.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.21.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.21.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.21.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+ "model.layers.22.input_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.22.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.input_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.23.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.input_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.24.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.input_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.25.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.26.input_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.26.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.26.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.26.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.26.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.26.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.26.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.26.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.26.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+ "model.layers.27.input_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.27.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.input_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.28.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.input_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.29.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.3.input_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.3.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.3.mlp.up_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.3.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+ "model.layers.30.input_layernorm.weight": "model-00008-of-00008.safetensors",
+ "model.layers.30.mlp.down_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.30.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.30.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.30.post_attention_layernorm.weight": "model-00008-of-00008.safetensors",
+ "model.layers.30.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.30.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.30.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.30.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+ "model.layers.31.input_layernorm.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.mlp.down_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.mlp.gate_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.mlp.up_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.post_attention_layernorm.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.self_attn.k_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.self_attn.o_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.self_attn.q_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.31.self_attn.v_proj.weight": "model-00008-of-00008.safetensors",
+ "model.layers.4.input_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.mlp.gate_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.mlp.up_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.4.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.input_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.mlp.gate_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.mlp.up_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.5.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.input_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.mlp.gate_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.mlp.up_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.6.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.input_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.mlp.up_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.7.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.8.input_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.8.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.8.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.8.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.8.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+ "model.layers.9.input_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+ "model.layers.9.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+ "model.norm.weight": "model-00008-of-00008.safetensors"
+ }
+}
diff --git a/out_tensor/model.layers.0.mlp.down_proj.safetensors b/out_tensor/model.layers.0.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e2445eb8326473d9873d455caa0d2cf80e7a3138
--- /dev/null
+++ b/out_tensor/model.layers.0.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d1983e86d69750adc6e8dfa80ec58a21264489c3d0c739ac2dbdf0c04bc5ea3
+size 20049048
diff --git a/out_tensor/model.layers.0.mlp.gate_proj.safetensors b/out_tensor/model.layers.0.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a6c5eb842f07b1b719ceb97bab14d0a0750e3ca8
--- /dev/null
+++ b/out_tensor/model.layers.0.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60ccaa791f7ac9d35b367458692f4b7316aaa564808ec874625bc1cc3964d6d4
+size 22955288
diff --git a/out_tensor/model.layers.0.mlp.up_proj.safetensors b/out_tensor/model.layers.0.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7988d0dc39734209e2f0084a1f3e537efaaa1dd5
--- /dev/null
+++ b/out_tensor/model.layers.0.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d25098a46a2f7eac5b7732611386670086d0206ed530adfa06dd4c09c1cc688b
+size 26854280
diff --git a/out_tensor/model.layers.0.self_attn.k_proj.safetensors b/out_tensor/model.layers.0.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4c2070da477be6f42cf9489367ee25ba278b29d3
--- /dev/null
+++ b/out_tensor/model.layers.0.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f962a6d27245894d2ed152add3517744f4bb444ea798a1e7f49528416d704915
+size 1164576
diff --git a/out_tensor/model.layers.0.self_attn.o_proj.safetensors b/out_tensor/model.layers.0.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e5285d993fe85d6f3cf7ddd020c41869254c4327
--- /dev/null
+++ b/out_tensor/model.layers.0.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:945b96148e665c2f211801ced07c9d5535980d94c2d73228a0963382858cfc30
+size 5522720
diff --git a/out_tensor/model.layers.0.self_attn.q_proj.safetensors b/out_tensor/model.layers.0.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..da99b643b2376b700558370c5fe77500f1ccc63c
--- /dev/null
+++ b/out_tensor/model.layers.0.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36f5a44316264ad4ed0229ce60c4faee41f1fe66916f371264df26fba2b469ae
+size 4605216
diff --git a/out_tensor/model.layers.0.self_attn.v_proj.safetensors b/out_tensor/model.layers.0.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..99ceac9d42a111257e5d8abee6ba7da2e18865f9
--- /dev/null
+++ b/out_tensor/model.layers.0.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60e7099141846823b5fb6fbd9d9031b0abe9dd56655062aed8e767bcb1d12e38
+size 1393952
diff --git a/out_tensor/model.layers.1.mlp.down_proj.safetensors b/out_tensor/model.layers.1.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7cd8dd685a29e8e58ab12aeacf35b676543c2fd2
--- /dev/null
+++ b/out_tensor/model.layers.1.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6637af0fc44d9e3da9d3bfb4fd48cc6da35766b577580419eb78ee4a53f2cee
+size 16051352
diff --git a/out_tensor/model.layers.1.mlp.gate_proj.safetensors b/out_tensor/model.layers.1.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..28eb642c5043cb5d69eb88dafe1bc4dc67e894df
--- /dev/null
+++ b/out_tensor/model.layers.1.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2db8d7864bf3af6ea3b346ee63e903f45d5a53f8e3de4a6218d84765801a83b3
+size 29606616
diff --git a/out_tensor/model.layers.1.mlp.up_proj.safetensors b/out_tensor/model.layers.1.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ebfb96388a19698b85983ce64cdcffa26424a100
--- /dev/null
+++ b/out_tensor/model.layers.1.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42280be30b8f3378a3e706fa58f92ed787200fb6661df5332abc8dc6c7d4479e
+size 29606600
diff --git a/out_tensor/model.layers.1.self_attn.k_proj.safetensors b/out_tensor/model.layers.1.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9dd5176a16e88b375d14a321a6a864c32415e9f1
--- /dev/null
+++ b/out_tensor/model.layers.1.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ddfc8a157c9b51f164ae2c37f9fb9df2d92f17d018e0eca861c1640418f874de
+size 1164576
diff --git a/out_tensor/model.layers.1.self_attn.o_proj.safetensors b/out_tensor/model.layers.1.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2ec2ef6dd95d7b3e17bdcc78fb8ec59dec3f91f9
--- /dev/null
+++ b/out_tensor/model.layers.1.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5ad17428f192abc2a1d259b9d4827bf2833e4a1e9ef1d7b9f5939a800aa47c5
+size 7423264
diff --git a/out_tensor/model.layers.1.self_attn.q_proj.safetensors b/out_tensor/model.layers.1.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..50f22b16bc594591a7c04a28b4bc96697a321c25
--- /dev/null
+++ b/out_tensor/model.layers.1.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3f66cef70804505bd5d10823d4bbe18ae82491e685b846027b3b6504eb253c1
+size 4605216
diff --git a/out_tensor/model.layers.1.self_attn.v_proj.safetensors b/out_tensor/model.layers.1.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d6782e25895110f6c33f774db2756b4db63948ac
--- /dev/null
+++ b/out_tensor/model.layers.1.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bc220e72208769a2b5f9d3c37377fe0d147841ac49077f5235848b2944998e6
+size 1869088
diff --git a/out_tensor/model.layers.10.mlp.down_proj.safetensors b/out_tensor/model.layers.10.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c5cce5e80852974b9a62eabb4655dbab42532e94
--- /dev/null
+++ b/out_tensor/model.layers.10.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa3a00b84a0aa667bb0cb98f67936e817e2efa758cb8c1fbfd77183d81409b0e
+size 30338208
diff --git a/out_tensor/model.layers.10.mlp.gate_proj.safetensors b/out_tensor/model.layers.10.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4b10f02c5117495cdfab5ea38f0ac6a56c6ede82
--- /dev/null
+++ b/out_tensor/model.layers.10.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c237b1d1e38398f2f654ec574d303fb730a53e7397649f21d10c86dafc17b70c
+size 29606616
diff --git a/out_tensor/model.layers.10.mlp.up_proj.safetensors b/out_tensor/model.layers.10.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..324da32265447a7eadcd6c3cb66e783b1fdf3a2c
--- /dev/null
+++ b/out_tensor/model.layers.10.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84bb56054223615066a5e4ebe741bd6f22493af6d717a9202326bcadb14031c3
+size 30295312
diff --git a/out_tensor/model.layers.10.self_attn.k_proj.safetensors b/out_tensor/model.layers.10.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..724b2ccf417b9a656c5165ecd702e9d205fdd325
--- /dev/null
+++ b/out_tensor/model.layers.10.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03f525c29d9de8aaaa2926f1db9806a8ea170b195c70e0ec6844c374f01143c4
+size 1443104
diff --git a/out_tensor/model.layers.10.self_attn.o_proj.safetensors b/out_tensor/model.layers.10.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fec1eeb7db4d9f1e8b2a55a6267d5f029de1460c
--- /dev/null
+++ b/out_tensor/model.layers.10.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48d0a028e977c10ec32ff2902f60ebf0213165285633cc53c693c7e745985ca0
+size 8668456
diff --git a/out_tensor/model.layers.10.self_attn.q_proj.safetensors b/out_tensor/model.layers.10.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dfbe75b7feed867f7108e38dd46e92cafc136876
--- /dev/null
+++ b/out_tensor/model.layers.10.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a7309bb2184ff9d54cd0c67b21e62a6e1dcfec0377ca8216e2ee8360343a648
+size 6374112
diff --git a/out_tensor/model.layers.10.self_attn.v_proj.safetensors b/out_tensor/model.layers.10.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bbddfaff4f17d6f23708e1358a647a8c89786e02
--- /dev/null
+++ b/out_tensor/model.layers.10.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfb41ceec5dfdc2a3c1c31e412c6142a210e6dcdf76064b2d79f96e065e4f6fc
+size 2130656
diff --git a/out_tensor/model.layers.11.mlp.down_proj.safetensors b/out_tensor/model.layers.11.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fe79862cb2e6a559c755eb2fb7e9d0fe1cf43912
--- /dev/null
+++ b/out_tensor/model.layers.11.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d543a52f4e4b40ff9f6227b236b2a004b2f52acb8c33fe1d30df7e8494ddcff
+size 31059104
diff --git a/out_tensor/model.layers.11.mlp.gate_proj.safetensors b/out_tensor/model.layers.11.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c00f4ffd400dc7ef6605aa471fd996103c8b16ac
--- /dev/null
+++ b/out_tensor/model.layers.11.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87e32ee466510d6ab403f8cc2d0af7df387f27caf4d2f7fb6a71a42dd934b7c4
+size 29606616
diff --git a/out_tensor/model.layers.11.mlp.up_proj.safetensors b/out_tensor/model.layers.11.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cbfc0ca8f8d645fd20f4253507208f383e06cf23
--- /dev/null
+++ b/out_tensor/model.layers.11.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:137348a44ac8a18806d7ad65e357fbd37daabe14b2214b8fb23d9e970a31bd2a
+size 30295312
diff --git a/out_tensor/model.layers.11.self_attn.k_proj.safetensors b/out_tensor/model.layers.11.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a687f749c9934ba299b07dbbde4a14fd1a1042e2
--- /dev/null
+++ b/out_tensor/model.layers.11.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d081fdc51ee72f90e67317bbc247295653993b52f6ef061ffa74e4d3e942adb
+size 1606368
diff --git a/out_tensor/model.layers.11.self_attn.o_proj.safetensors b/out_tensor/model.layers.11.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c94df344bcfb566ad5fe95c7c3848d655d0f98c6
--- /dev/null
+++ b/out_tensor/model.layers.11.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61399c8ee13597e9253816c116c94195689428bd2e2a54c0ad193bcefd9b8986
+size 8668456
diff --git a/out_tensor/model.layers.11.self_attn.q_proj.safetensors b/out_tensor/model.layers.11.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4f11de9457fcf3baa4070364897acb6ea985508f
--- /dev/null
+++ b/out_tensor/model.layers.11.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c9e4aedd433d1d66b5ad4b1b7c411681a1a61a0614aa5ebfde8f95e9eb235e5
+size 6571304
diff --git a/out_tensor/model.layers.11.self_attn.v_proj.safetensors b/out_tensor/model.layers.11.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..802786e531e4439254e6cc584eed6ff5cc73569c
--- /dev/null
+++ b/out_tensor/model.layers.11.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1f18defa8985c65ec0a700c7eddfd4b47237b5ca8840e203350c6378245f0ab
+size 2130656
diff --git a/out_tensor/model.layers.12.mlp.down_proj.safetensors b/out_tensor/model.layers.12.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..997a5972f70975b23703b0313dee7f759a1d2066
--- /dev/null
+++ b/out_tensor/model.layers.12.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d248304654a13e1e994d09651c86765e276627a131e281ba7acd2ac7d1ff304
+size 31059104
diff --git a/out_tensor/model.layers.12.mlp.gate_proj.safetensors b/out_tensor/model.layers.12.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e09c9e588cb986f96e301350bac2f6db7f05de62
--- /dev/null
+++ b/out_tensor/model.layers.12.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae30a9e702f9a95cb704d3ac4e8edc9712c658380d3e170fca79f4e86a685037
+size 29606616
diff --git a/out_tensor/model.layers.12.mlp.up_proj.safetensors b/out_tensor/model.layers.12.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a06ca498a132c615d78d9be12fde811fd254bcaa
--- /dev/null
+++ b/out_tensor/model.layers.12.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e043ef2ccf059c5397a139d22bb2ef4ec883a6a4661816d4a20c859cec7681a7
+size 30295312
diff --git a/out_tensor/model.layers.12.self_attn.k_proj.safetensors b/out_tensor/model.layers.12.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3bdeb42f287f2231bbb0310933f2837bac6810bd
--- /dev/null
+++ b/out_tensor/model.layers.12.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83b118eb4581030eb6c0c9236afdaf4b1fcee5cce77cdfd855a19f44473fff42
+size 1606368
diff --git a/out_tensor/model.layers.12.self_attn.o_proj.safetensors b/out_tensor/model.layers.12.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2e4677df5f0af9cfdda928e04fd25cfec07ad725
--- /dev/null
+++ b/out_tensor/model.layers.12.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:687c9d75c30dab3b3b0f848cb8ba932e11cf9c6ba0821e1766dd7df087879020
+size 8668456
diff --git a/out_tensor/model.layers.12.self_attn.q_proj.safetensors b/out_tensor/model.layers.12.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b28514fa9e97369324ca1115076c2db33a631cfa
--- /dev/null
+++ b/out_tensor/model.layers.12.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f01ceda5c88cb1fd102420da6f725400714b495a8e9ad5ab7815c422f64f9c90
+size 6571304
diff --git a/out_tensor/model.layers.12.self_attn.v_proj.safetensors b/out_tensor/model.layers.12.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..966f3ca8150cc4497adb19700e31016eaa682103
--- /dev/null
+++ b/out_tensor/model.layers.12.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50ebcc4b32bbea5db0ba7cb3ead916ad88e5cafdd767c130ac5519c7fccc6629
+size 2130656
diff --git a/out_tensor/model.layers.13.mlp.down_proj.safetensors b/out_tensor/model.layers.13.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..56677569f19cd5f4e1152324ba44dfa2f6321bb6
--- /dev/null
+++ b/out_tensor/model.layers.13.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9a55d73f3e9dac0631194a35090a23ad3d00bd156e7f0bb043b7036e4d22498
+size 31780000
diff --git a/out_tensor/model.layers.13.mlp.gate_proj.safetensors b/out_tensor/model.layers.13.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c0e218b185fe78509c62c1e2ef2fa2721f4c58f2
--- /dev/null
+++ b/out_tensor/model.layers.13.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8d4a3db6426a96df90fa5fb5237135fb089c2357c3e8734c40099a51d43df24
+size 29606616
diff --git a/out_tensor/model.layers.13.mlp.up_proj.safetensors b/out_tensor/model.layers.13.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..50b7da51badf3e9fdacbab22cd8ff42a6906f7b0
--- /dev/null
+++ b/out_tensor/model.layers.13.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d44c428faaeb8995cd8528a6cba0aa2943af1904deb58fb2ed5b4d8b359a712
+size 30983440
diff --git a/out_tensor/model.layers.13.self_attn.k_proj.safetensors b/out_tensor/model.layers.13.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..731c8875ac1a708517b656f691f49e373e6e495b
--- /dev/null
+++ b/out_tensor/model.layers.13.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ad3873c4756b83830ebc0202d596f99fe9f972b0633e7745636e94f4fe4fdeb
+size 1606368
diff --git a/out_tensor/model.layers.13.self_attn.o_proj.safetensors b/out_tensor/model.layers.13.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..da23cffcb29b57e307fc8b2e4411a27ee9e0b150
--- /dev/null
+++ b/out_tensor/model.layers.13.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e53c8e0aee98efb2dd008bb23e09c2471e4c9b02dd34fe42b44b242f14a1a61
+size 8865064
diff --git a/out_tensor/model.layers.13.self_attn.q_proj.safetensors b/out_tensor/model.layers.13.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3242ab95614b687f409a6ace497553b01328e511
--- /dev/null
+++ b/out_tensor/model.layers.13.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d99b5b61d97eb836fdefbb5d67fb7da292302efdbe745ae9fac3954aab3ffe2
+size 6571304
diff --git a/out_tensor/model.layers.13.self_attn.v_proj.safetensors b/out_tensor/model.layers.13.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..504a338382b8cb3e9fb989d466ce1356a6e4e73a
--- /dev/null
+++ b/out_tensor/model.layers.13.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1bba7557ece2ec72765703862238b8b6a585b013be3585b0731fcef730a4949
+size 2180384
diff --git a/out_tensor/model.layers.14.mlp.down_proj.safetensors b/out_tensor/model.layers.14.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bbe5003bc1829899ef1a9f2d216428fbbe017570
--- /dev/null
+++ b/out_tensor/model.layers.14.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ab739601a698492ab2a576527a1b963a92143910b9662496d0e7ec3ee46f133
+size 31059104
diff --git a/out_tensor/model.layers.14.mlp.gate_proj.safetensors b/out_tensor/model.layers.14.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c449f67c12b21d71cdb92d56d7a5b5dfc2e2ccf2
--- /dev/null
+++ b/out_tensor/model.layers.14.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b26b02fdc4650a4dc25525393a840721434c7acc963e7fad425d0155ab6d7b1
+size 29606616
diff --git a/out_tensor/model.layers.14.mlp.up_proj.safetensors b/out_tensor/model.layers.14.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7682df81557edc1ebb1481df23db2cf43479ff46
--- /dev/null
+++ b/out_tensor/model.layers.14.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8886c374c89f7dd4587ab91f27a9eaf5bcdb362deae187feeb8317b004ef5897
+size 30983440
diff --git a/out_tensor/model.layers.14.self_attn.k_proj.safetensors b/out_tensor/model.layers.14.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d7bd3cc63e877c98ecf56b9c7409529a3e318fb8
--- /dev/null
+++ b/out_tensor/model.layers.14.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3adffaf8350398455dcfc23e29f233bc2fe30f843b3f166413b4747fda2d5de
+size 1606368
diff --git a/out_tensor/model.layers.14.self_attn.o_proj.safetensors b/out_tensor/model.layers.14.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e9ca98f00959f798186fbe8a15aef7884bf7d85f
--- /dev/null
+++ b/out_tensor/model.layers.14.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8a50b822428643e0ed4c337c493dbb408f585947993633b4ee95ddac54640a8
+size 10568424
diff --git a/out_tensor/model.layers.14.self_attn.q_proj.safetensors b/out_tensor/model.layers.14.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..75461747084f2c34ceb52bfed026f5c4b2ae933a
--- /dev/null
+++ b/out_tensor/model.layers.14.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34d347aa94b996c2ebdc877c28ead60270cd3d2cbb5f48253f198eebe2329932
+size 6702376
diff --git a/out_tensor/model.layers.14.self_attn.v_proj.safetensors b/out_tensor/model.layers.14.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..681c5e85a1300666563e3b9dfc777760e5eda527
--- /dev/null
+++ b/out_tensor/model.layers.14.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:317491e08e4508008776a798cb64e27dc5ee320e305a0a8ee4d762b3202a29c1
+size 2180384
diff --git a/out_tensor/model.layers.15.mlp.down_proj.safetensors b/out_tensor/model.layers.15.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fce270e283aaa26fbdd2a1d2dceb0ef588a2fa9e
--- /dev/null
+++ b/out_tensor/model.layers.15.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecc3a56184ff4a1d8f583a70c970fb7a77fcff10ddd0aa00feba7e0f9037b15c
+size 30338208
diff --git a/out_tensor/model.layers.15.mlp.gate_proj.safetensors b/out_tensor/model.layers.15.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1caf38e3190c3070d4e6f9037832b1a4687b69af
--- /dev/null
+++ b/out_tensor/model.layers.15.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c160a44fd2920fb1fb439280af9cd3fbc6c7aa83f25d93d0405094d68d42ae70
+size 29606616
diff --git a/out_tensor/model.layers.15.mlp.up_proj.safetensors b/out_tensor/model.layers.15.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8485e1a9db0c2b9e7513e011f8535f15b0567002
--- /dev/null
+++ b/out_tensor/model.layers.15.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f39d5c4bc4bc3ae43027717249f68ff6364011ab5b84040ba8634c0cee0ceb30
+size 30983440
diff --git a/out_tensor/model.layers.15.self_attn.k_proj.safetensors b/out_tensor/model.layers.15.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..32ab7a270a89e82a4ce18749a5f23df17c22349f
--- /dev/null
+++ b/out_tensor/model.layers.15.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58907d4cee279bd4502c842d73541260602f00377d56cf9bb5a54c9ac31e4957
+size 1656096
diff --git a/out_tensor/model.layers.15.self_attn.o_proj.safetensors b/out_tensor/model.layers.15.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f4502489400153ed9193eb032e98ebe9cae53e84
--- /dev/null
+++ b/out_tensor/model.layers.15.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:510e0e235e85f6999c3728d7783322fd55ce6e1b597cfefe034d7eedc1452263
+size 8668456
diff --git a/out_tensor/model.layers.15.self_attn.q_proj.safetensors b/out_tensor/model.layers.15.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6c38ee7f0f3e1b05b22ba92ea05e5a401811e361
--- /dev/null
+++ b/out_tensor/model.layers.15.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe5f3c4cd07ac9a984e5fddbb7add7dd5a5001c8e1ec734bf0531239ab9ebbf0
+size 7423272
diff --git a/out_tensor/model.layers.15.self_attn.v_proj.safetensors b/out_tensor/model.layers.15.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..10e94a58f1fe8a9694eb77257445a519bda01379
--- /dev/null
+++ b/out_tensor/model.layers.15.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cdfd3545a03b4dfd30df5a2b75154a587a61039cf7248bff4083abefe8431ea
+size 2229536
diff --git a/out_tensor/model.layers.16.mlp.down_proj.safetensors b/out_tensor/model.layers.16.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65ac2f4a0a74de463b4a684a645da672f0b73407
--- /dev/null
+++ b/out_tensor/model.layers.16.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a67b0cb128353c9700940d48a581ecee1becaf94b6cf09f906130de82294b92
+size 30338208
diff --git a/out_tensor/model.layers.16.mlp.gate_proj.safetensors b/out_tensor/model.layers.16.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6d874214a28cc249fb1e03ae286f3d8a22e75f1a
--- /dev/null
+++ b/out_tensor/model.layers.16.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bda8693cf230d958b8844ab5d097fe2060d1bdf9eff74ce21d9d0fc82086cac
+size 29606616
diff --git a/out_tensor/model.layers.16.mlp.up_proj.safetensors b/out_tensor/model.layers.16.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b78c494ae71c420c579729ac71aea2e57655fab9
--- /dev/null
+++ b/out_tensor/model.layers.16.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f657383337be8141784b39f52f50271c3f7fdaf42a1439f00823112f7da9f6d8
+size 30295312
diff --git a/out_tensor/model.layers.16.self_attn.k_proj.safetensors b/out_tensor/model.layers.16.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b552d624a055632d61eb8a3aa0d1da5e0983e05a
--- /dev/null
+++ b/out_tensor/model.layers.16.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:231233a87df9c093349c51496af9c9d9edd11ca1caa6e6a01b47bf9bf93cbfc1
+size 1656096
diff --git a/out_tensor/model.layers.16.self_attn.o_proj.safetensors b/out_tensor/model.layers.16.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d83df8912aa880c02f4ef4c540917555a71ae039
--- /dev/null
+++ b/out_tensor/model.layers.16.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d70286839595651ca9f892ad6cc0a34ba99d32ff51631042ff7c0ebb443f7db4
+size 8865064
diff --git a/out_tensor/model.layers.16.self_attn.q_proj.safetensors b/out_tensor/model.layers.16.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1b7a9828475042c065a50ede27a0f69e8f35726b
--- /dev/null
+++ b/out_tensor/model.layers.16.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:102346426b379f13c4ed55df0f249b4f065d519be35a4ada9475e24fb6ea3fc9
+size 6767912
diff --git a/out_tensor/model.layers.16.self_attn.v_proj.safetensors b/out_tensor/model.layers.16.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..644b0f74cd9f0a7346293ef3c021a5287ea5e8ad
--- /dev/null
+++ b/out_tensor/model.layers.16.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd2b452a07e38c97a5ef6ab7fa0d5f28104ffc361d689418912f67832761ebaf
+size 2229536
diff --git a/out_tensor/model.layers.17.mlp.down_proj.safetensors b/out_tensor/model.layers.17.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..81b74c38b96127842e5d8b323ff67e1639d899fc
--- /dev/null
+++ b/out_tensor/model.layers.17.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37368d22ccc1667fbe8b32f00b4dd9cd709b785e2fe867bde4f519a073f2e20b
+size 31059104
diff --git a/out_tensor/model.layers.17.mlp.gate_proj.safetensors b/out_tensor/model.layers.17.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4a8c3d1ce6a3382833ae579c145834b299b6f6f1
--- /dev/null
+++ b/out_tensor/model.layers.17.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13d846996b88b764f9bf6486c421b647b4e79b40b07e977811048e5c77eebd03
+size 29606616
diff --git a/out_tensor/model.layers.17.mlp.up_proj.safetensors b/out_tensor/model.layers.17.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..07b0d30e44a3211a818e26bd1f640fea82b10838
--- /dev/null
+++ b/out_tensor/model.layers.17.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466c114b8d92ac83ee74d81ab86d0e2457300d712a628a47c47e3b74a74c8d43
+size 30983440
diff --git a/out_tensor/model.layers.17.self_attn.k_proj.safetensors b/out_tensor/model.layers.17.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..400a7c2d4acb0d9380fad60cff277ce68df3ca52
--- /dev/null
+++ b/out_tensor/model.layers.17.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be74b9b67c532868a764c60c50448993fa273f4f6d380f4ccf36f6241e574a32
+size 1606368
diff --git a/out_tensor/model.layers.17.self_attn.o_proj.safetensors b/out_tensor/model.layers.17.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fbb688bba4dbeae1566d14bd60d62b2eb9286350
--- /dev/null
+++ b/out_tensor/model.layers.17.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f301dfbae11990a4491b8517b72485183a0f4271f3abe718ce9ba6f69ddc208
+size 8668456
diff --git a/out_tensor/model.layers.17.self_attn.q_proj.safetensors b/out_tensor/model.layers.17.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ca4fc03910ae0fdd58b92559a21d80b72f6e7804
--- /dev/null
+++ b/out_tensor/model.layers.17.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c780a66765e80d6b0f42f49b710031dbd41ba602d4f088d6e56a99fa11d2d81
+size 6571304
diff --git a/out_tensor/model.layers.17.self_attn.v_proj.safetensors b/out_tensor/model.layers.17.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ad9cae2601177d3c479b0804fee830cc69222afc
--- /dev/null
+++ b/out_tensor/model.layers.17.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1d4df1f50f110de88d312ba0daa4f3c68a069bc8b85e2a79abc104044a86d77
+size 2180384
diff --git a/out_tensor/model.layers.18.mlp.down_proj.safetensors b/out_tensor/model.layers.18.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..595f063d23df65b77e0dfbf45ad7ccf254ba6023
--- /dev/null
+++ b/out_tensor/model.layers.18.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13ff96611185904fea79a988d2c112d2f235c4ddea239e0750718fcf2a9b44a6
+size 30338208
diff --git a/out_tensor/model.layers.18.mlp.gate_proj.safetensors b/out_tensor/model.layers.18.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ad1415372aabbeae8f90118c4797553cd35c19a8
--- /dev/null
+++ b/out_tensor/model.layers.18.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56720c5efc528d65c10b3f86b46b9285674956ae15d7527e3a2d1e1b796a801b
+size 29606616
diff --git a/out_tensor/model.layers.18.mlp.up_proj.safetensors b/out_tensor/model.layers.18.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..96f6ab5711e4b9f689d7daef770ac93d9a3cce69
--- /dev/null
+++ b/out_tensor/model.layers.18.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48526bf72a216a1a0f460b75925bda10242cf7f4ba743ddf8ece4e882aa3edf4
+size 30295312
diff --git a/out_tensor/model.layers.18.self_attn.k_proj.safetensors b/out_tensor/model.layers.18.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ab34b819f99f087c90a07ad2e935b40d7812422e
--- /dev/null
+++ b/out_tensor/model.layers.18.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be86cb1d2377c39600e36d5dc331fc16fa2e7e4123cb04d50be758a5856c7f95
+size 1656096
diff --git a/out_tensor/model.layers.18.self_attn.o_proj.safetensors b/out_tensor/model.layers.18.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2beeb6340ae7907d7cc2ab98b126246c4f20187d
--- /dev/null
+++ b/out_tensor/model.layers.18.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:321288e0db1ff56d7c5e1ff44d1aec4b82524cd3bb417ce05677d9c98024db1e
+size 8668456
diff --git a/out_tensor/model.layers.18.self_attn.q_proj.safetensors b/out_tensor/model.layers.18.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..88a43bc91f64717a814b4e0ab5910b6dfc87ac4d
--- /dev/null
+++ b/out_tensor/model.layers.18.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b58f5ce3ff97f39576cd8ab5faff0bfdf84664c26b0310cee71adcc65792a868
+size 6702376
diff --git a/out_tensor/model.layers.18.self_attn.v_proj.safetensors b/out_tensor/model.layers.18.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..340c084846a0808e318f476bbde4fe8715ac1a89
--- /dev/null
+++ b/out_tensor/model.layers.18.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b421e8a93c54801089089b79bc700c470e90df05ad37c7aa2542c14beeae9c2
+size 2180384
diff --git a/out_tensor/model.layers.19.mlp.down_proj.safetensors b/out_tensor/model.layers.19.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fee947d100f08e9e007c60692f40171962bb1fcd
--- /dev/null
+++ b/out_tensor/model.layers.19.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0004152f2458bff6ad2f8d526dc8d4c6e08b9af03cbc7406e5a36bc2bea74a6d
+size 30338208
diff --git a/out_tensor/model.layers.19.mlp.gate_proj.safetensors b/out_tensor/model.layers.19.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..456abae44030011270ac88738c63786f36380b96
--- /dev/null
+++ b/out_tensor/model.layers.19.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32208859150505a661611ffabb383ef8eb52e9b61cbf0712028371912dc5b84f
+size 29606616
diff --git a/out_tensor/model.layers.19.mlp.up_proj.safetensors b/out_tensor/model.layers.19.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4d263b529d986b27976513075ffcb27dd5ac4aa6
--- /dev/null
+++ b/out_tensor/model.layers.19.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76a1cb45f56d1dce6eec050556bd0e31aef9be710c92357241c169649181c037
+size 30295312
diff --git a/out_tensor/model.layers.19.self_attn.k_proj.safetensors b/out_tensor/model.layers.19.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..27264f907d7e6d974d7c5007690fbab78bf69424
--- /dev/null
+++ b/out_tensor/model.layers.19.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48c9ba9aa17be2019b3be7350a8ce2bce424a7e73841fa7aa08fd25d3496320f
+size 1656096
diff --git a/out_tensor/model.layers.19.self_attn.o_proj.safetensors b/out_tensor/model.layers.19.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8a5106e37cdfdecadb2861a06182fd6cc954bf41
--- /dev/null
+++ b/out_tensor/model.layers.19.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebe5e7cd34f6532dc921c742044a54d62a407ef0936c60af5ccc70ad38a7ec33
+size 8668456
diff --git a/out_tensor/model.layers.19.self_attn.q_proj.safetensors b/out_tensor/model.layers.19.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3cf3465bd45d41aaa6aa3c16bd019bb3d1a5e9a9
--- /dev/null
+++ b/out_tensor/model.layers.19.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aedc2dcaed6c60c30a3152af03483b90d4747713475d975822674f0b2c94b683
+size 6767912
diff --git a/out_tensor/model.layers.19.self_attn.v_proj.safetensors b/out_tensor/model.layers.19.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2e509b4cb726732ec675cce62f1b6c7b3b81dd83
--- /dev/null
+++ b/out_tensor/model.layers.19.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d8d693b7427e5916f439a548b341731aca02d7fa3592158c3c3d46797cedc96
+size 2229536
diff --git a/out_tensor/model.layers.2.mlp.down_proj.safetensors b/out_tensor/model.layers.2.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65dc42ecf2a5abc74fb37f70cc97da668f480c68
--- /dev/null
+++ b/out_tensor/model.layers.2.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fdf2ce280cf77d6ba2765d7f83e538f1307ca56e6106c4e0897ef69198235fb
+size 29648056
diff --git a/out_tensor/model.layers.2.mlp.gate_proj.safetensors b/out_tensor/model.layers.2.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fd98c04325fc59efd54018bf971e8a3c61c51a33
--- /dev/null
+++ b/out_tensor/model.layers.2.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ee1886398744769f935856e9bf15e809cffa6abadad9efb27f6ad50f94c867c
+size 29606616
diff --git a/out_tensor/model.layers.2.mlp.up_proj.safetensors b/out_tensor/model.layers.2.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ddf6cb3c4cf5385c4c2360a9ec0f73d7b3c0e9b6
--- /dev/null
+++ b/out_tensor/model.layers.2.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:197532904fa120ef3ed5516a2cc771520a48f9868baedb7272468a2519fadada
+size 30295312
diff --git a/out_tensor/model.layers.2.self_attn.k_proj.safetensors b/out_tensor/model.layers.2.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..67d2bf9a1b064912f31a9dd1dbb69c38e993d376
--- /dev/null
+++ b/out_tensor/model.layers.2.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15459ea712103d4e9a13a17055fe494505d5906bbeacb09e1ff60ed360e4ba9e
+size 1262880
diff --git a/out_tensor/model.layers.2.self_attn.o_proj.safetensors b/out_tensor/model.layers.2.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a9ec067b0bb435a4ce5db2d56967e48b0d9b0b64
--- /dev/null
+++ b/out_tensor/model.layers.2.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0dec8e3aa46d27799fca34c0d55171bb6b02949644565ffdc0f76f388326f40
+size 8471256
diff --git a/out_tensor/model.layers.2.self_attn.q_proj.safetensors b/out_tensor/model.layers.2.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7b70794a76579dd08432eb338a17fd78f6b2e0a0
--- /dev/null
+++ b/out_tensor/model.layers.2.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94e747f9240d1d873cea73e1e51ebb364731cf4f38e856b2400282a6d9c249e2
+size 4998432
diff --git a/out_tensor/model.layers.2.self_attn.v_proj.safetensors b/out_tensor/model.layers.2.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ca2277c393588312c65dcdcc72117d0992304a51
--- /dev/null
+++ b/out_tensor/model.layers.2.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b435b94e0eb00dadf58b99a74fdbbadc02128e720abd153419e1f691f037a08f
+size 2130648
diff --git a/out_tensor/model.layers.20.mlp.down_proj.safetensors b/out_tensor/model.layers.20.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d0a15864d3c2cc129c726bcf7ede3236615ee47f
--- /dev/null
+++ b/out_tensor/model.layers.20.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeaf60ddd06959e11973e0988cf6e20241ef09d0e89036bb8a7ad933069d41ef
+size 30338208
diff --git a/out_tensor/model.layers.20.mlp.gate_proj.safetensors b/out_tensor/model.layers.20.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65a3f6b63fe7d38138e4e6371771995b566983b4
--- /dev/null
+++ b/out_tensor/model.layers.20.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de38c262fedc355bdc52582e0d9086010c5c2e9c85847690b6753e4e77e55757
+size 29606616
diff --git a/out_tensor/model.layers.20.mlp.up_proj.safetensors b/out_tensor/model.layers.20.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a48f58cdc76b4f508249538aee9d790a74ece0b3
--- /dev/null
+++ b/out_tensor/model.layers.20.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f385300653581f95e3f0f3b9d32cb683c2078939b8f8ec28660e9473f0528d46
+size 30295312
diff --git a/out_tensor/model.layers.20.self_attn.k_proj.safetensors b/out_tensor/model.layers.20.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4823efbb734e3af837500c64795bf40327b2712a
--- /dev/null
+++ b/out_tensor/model.layers.20.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93b1ea6e503003e9eddb8fc26d715b7c47ce8c99c0cb67ba51d6fb687c9bddb9
+size 1656096
diff --git a/out_tensor/model.layers.20.self_attn.o_proj.safetensors b/out_tensor/model.layers.20.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ef9516c9f1aa2b342528689c7f7a9b63f8586ae8
--- /dev/null
+++ b/out_tensor/model.layers.20.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eabab48054b109ef4707c9f62b30d30d5c5786a2e25386c0c285a6bf8c6630f6
+size 8865064
diff --git a/out_tensor/model.layers.20.self_attn.q_proj.safetensors b/out_tensor/model.layers.20.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..435cf25bbca63223e8402e3f4ea72742e3487a48
--- /dev/null
+++ b/out_tensor/model.layers.20.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c736eb87eec80236bd2865bfd17c42d1b8069c4777c1fbbdf682544fcf94d3ac
+size 6767912
diff --git a/out_tensor/model.layers.20.self_attn.v_proj.safetensors b/out_tensor/model.layers.20.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65f1c29ef722343211352eb16727cfa26f88efdb
--- /dev/null
+++ b/out_tensor/model.layers.20.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0c72ed0e541ec5c904993af335394ec3d6baea5f093f6247bc0cb3637a0cf68
+size 2229536
diff --git a/out_tensor/model.layers.21.mlp.down_proj.safetensors b/out_tensor/model.layers.21.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b04d59cd27c32511ab598428fb1f6267d3e43d8e
--- /dev/null
+++ b/out_tensor/model.layers.21.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d899b5deb5c0f517e389fa017c4542d802d909d2a097b45173e1455f4d09166
+size 30338208
diff --git a/out_tensor/model.layers.21.mlp.gate_proj.safetensors b/out_tensor/model.layers.21.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f0dd86bc7ae3c8dfcd2b959bd36704cf5ab8f818
--- /dev/null
+++ b/out_tensor/model.layers.21.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfa3df7d7c9303586081b2dac1e711631ecb33b36ee76b0a6e9f24d05a6fde8d
+size 29606616
diff --git a/out_tensor/model.layers.21.mlp.up_proj.safetensors b/out_tensor/model.layers.21.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..425aab01e795e5dd48823e4fb152dca501bb03ee
--- /dev/null
+++ b/out_tensor/model.layers.21.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a2decb10da26edb0910525eb3a0bd7370d0a5222b74e68b3f61a9dddb0c4dd7
+size 30983440
diff --git a/out_tensor/model.layers.21.self_attn.k_proj.safetensors b/out_tensor/model.layers.21.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..01acc57137c1a3fe92e3963f886acb300828ace0
--- /dev/null
+++ b/out_tensor/model.layers.21.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70e07d2faf6eac06c77fe8b926bdbc32e2defa45d3b4754aa393a98620540dce
+size 1656096
diff --git a/out_tensor/model.layers.21.self_attn.o_proj.safetensors b/out_tensor/model.layers.21.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..76da42762269205182b5932c5e912f6650074b60
--- /dev/null
+++ b/out_tensor/model.layers.21.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97e0d1c22506ee6121b87b6be22e58549610cc6e4c970852dd15e8f1a91f836c
+size 8865064
diff --git a/out_tensor/model.layers.21.self_attn.q_proj.safetensors b/out_tensor/model.layers.21.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8acfe21415ebfb906402814a6026a2f658bf3314
--- /dev/null
+++ b/out_tensor/model.layers.21.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e513d8dadce3014555417cc951468c4f7acf69ce7d67448dee0e4efb8f8b3b9f
+size 6702376
diff --git a/out_tensor/model.layers.21.self_attn.v_proj.safetensors b/out_tensor/model.layers.21.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4646e455e67fcccdc10f345791f7fd5384da59ee
--- /dev/null
+++ b/out_tensor/model.layers.21.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd5ac896c199c07246bccd854c694a12876978878c5679beffe5308eb197ec4f
+size 2180384
diff --git a/out_tensor/model.layers.22.mlp.down_proj.safetensors b/out_tensor/model.layers.22.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a3ae58ca6e87ffe99fd466dbe0cfd639dfb2f7a5
--- /dev/null
+++ b/out_tensor/model.layers.22.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2c2d22608fada0e7e7885929d4f391adc9c45e9fdb92201873758171d96e978
+size 30338208
diff --git a/out_tensor/model.layers.22.mlp.gate_proj.safetensors b/out_tensor/model.layers.22.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b72efe2cc6a298a330e30a5634201f733f5daabf
--- /dev/null
+++ b/out_tensor/model.layers.22.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93dd23c808b5ee745fe59eed84900e3915335b28a5e1a4b257d62ac6e6fa5bf6
+size 29606616
diff --git a/out_tensor/model.layers.22.mlp.up_proj.safetensors b/out_tensor/model.layers.22.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..86b3d9de0b0f00ded899aca1a5b3b4ef68a4d5e0
--- /dev/null
+++ b/out_tensor/model.layers.22.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2a363130d72402fa118d252f0f765767769088182ed6898bcf9a3a9bda7527c
+size 31671568
diff --git a/out_tensor/model.layers.22.self_attn.k_proj.safetensors b/out_tensor/model.layers.22.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f281bc88179a89277f1408889e38055669bda2f5
--- /dev/null
+++ b/out_tensor/model.layers.22.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:678be0209d18463af530e423ce18478668ef1322bee977be93cd78064ca4f6a3
+size 1688864
diff --git a/out_tensor/model.layers.22.self_attn.o_proj.safetensors b/out_tensor/model.layers.22.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6856e4ef379d80e9e2745bb5fac3723ae5ad071a
--- /dev/null
+++ b/out_tensor/model.layers.22.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d393ff5e88dc08b6e488d0aefaec26b41293d1b61b7603e7fac6bf734599919
+size 8668456
diff --git a/out_tensor/model.layers.22.self_attn.q_proj.safetensors b/out_tensor/model.layers.22.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..96b6d732d72dd29b11d880a37b51073da4d6c17f
--- /dev/null
+++ b/out_tensor/model.layers.22.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d35c99b3a37b7910db7d444516a3ee01af7caf6a74023217e2a2f5d6ec5aa845
+size 6702376
diff --git a/out_tensor/model.layers.22.self_attn.v_proj.safetensors b/out_tensor/model.layers.22.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b9459b8d03215e65c947984c38b88f21cabbd0cf
--- /dev/null
+++ b/out_tensor/model.layers.22.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:042e4cc1ef2176723b22197e4c77f2f69fa74b770c05ffeea674e89f2763009f
+size 2180384
diff --git a/out_tensor/model.layers.23.mlp.down_proj.safetensors b/out_tensor/model.layers.23.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..de893a159aca5921f0a9e749528af12af9511550
--- /dev/null
+++ b/out_tensor/model.layers.23.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9657ceb917ec57ba742734faf781b1ee386171c53e33822a71dd7b9d9bd2de14
+size 30338208
diff --git a/out_tensor/model.layers.23.mlp.gate_proj.safetensors b/out_tensor/model.layers.23.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f6ebb9a55c3cfd3b254a2da2176d8b24fec473c7
--- /dev/null
+++ b/out_tensor/model.layers.23.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:793b97b41b6896a7425ef2048921d003ed282d3ed87781d90da4cb0bc74c344d
+size 29606616
diff --git a/out_tensor/model.layers.23.mlp.up_proj.safetensors b/out_tensor/model.layers.23.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3f4357b6af4ed50fa64c5a2af6fe383704dfc56f
--- /dev/null
+++ b/out_tensor/model.layers.23.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea1e99832e39323617bd29401f50b75bd9c999516ccc0b017e5c955f67e58562
+size 36946640
diff --git a/out_tensor/model.layers.23.self_attn.k_proj.safetensors b/out_tensor/model.layers.23.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9340a9bdcbe0aa4f9d5ee86344b0f6398e177ed4
--- /dev/null
+++ b/out_tensor/model.layers.23.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cbf85186d9bffc3c48facc3696cc2dabcd098cf0d5b9a845c1f8387bca82451
+size 1688864
diff --git a/out_tensor/model.layers.23.self_attn.o_proj.safetensors b/out_tensor/model.layers.23.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8f24717d51c2d670586dad737a9441191b7c6048
--- /dev/null
+++ b/out_tensor/model.layers.23.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c74cf20b63f9735d1b48a9c651068ef0e0d4fc7bd195254579baa78f6d68af5d
+size 8668456
diff --git a/out_tensor/model.layers.23.self_attn.q_proj.safetensors b/out_tensor/model.layers.23.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e48f9c2d4af3a8920e506d5a8929ad0ff52986f3
--- /dev/null
+++ b/out_tensor/model.layers.23.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a233a84e2c6eb9a659254f586563695076402fd00c71af061c26dc58eb9c8bad
+size 6702376
diff --git a/out_tensor/model.layers.23.self_attn.v_proj.safetensors b/out_tensor/model.layers.23.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b013222967b6941d607011a812108c9c566ce27c
--- /dev/null
+++ b/out_tensor/model.layers.23.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27beca4de04e01a27ab395107f7b16b6020ded5cf7d55da81f0194078e2a3f8a
+size 2229536
diff --git a/out_tensor/model.layers.24.mlp.down_proj.safetensors b/out_tensor/model.layers.24.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cfbf84414d11e0255c0ab3d79ae3424cb1558dea
--- /dev/null
+++ b/out_tensor/model.layers.24.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce17129c67fdc66507d8d1b1d90e07fcf45e0d95df24eec49e15055a08c97039
+size 30338208
diff --git a/out_tensor/model.layers.24.mlp.gate_proj.safetensors b/out_tensor/model.layers.24.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..673554deb2415f95eada4b057dbb271be43214a4
--- /dev/null
+++ b/out_tensor/model.layers.24.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65875557b83b524a301774660c036c258971aeef3119e9ca8ce9f5eeaf3ef93f
+size 29606616
diff --git a/out_tensor/model.layers.24.mlp.up_proj.safetensors b/out_tensor/model.layers.24.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d12d8e2a2614cb0f4e332d7496e8c26d51e3661b
--- /dev/null
+++ b/out_tensor/model.layers.24.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94540a715277d8a4fd62204e24e2eb9aea16ec2d9ec75e06a2235efeddef16c9
+size 36946640
diff --git a/out_tensor/model.layers.24.self_attn.k_proj.safetensors b/out_tensor/model.layers.24.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..17525b378f035bd1cb40d54db504183d053340f0
--- /dev/null
+++ b/out_tensor/model.layers.24.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44b3a6d061c13dd264687372ebe75052cfd52bc5b635c6070a4c8c34836579b0
+size 1688864
diff --git a/out_tensor/model.layers.24.self_attn.o_proj.safetensors b/out_tensor/model.layers.24.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f2fb7c03942810fae2d380be03a1b16f597eadcc
--- /dev/null
+++ b/out_tensor/model.layers.24.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc56366cf0bcf3ec628fc931564c8c831b442fddf88ec91dbfccb0b8cdaa1ec9
+size 8471264
diff --git a/out_tensor/model.layers.24.self_attn.q_proj.safetensors b/out_tensor/model.layers.24.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c8edb5902c1d8eb596f6f0845e4a21a350df220b
--- /dev/null
+++ b/out_tensor/model.layers.24.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f3a09071d9d11a4a54f60c45255c448a38a2e429d2a24ba2c7f77d8636516f6
+size 7423272
diff --git a/out_tensor/model.layers.24.self_attn.v_proj.safetensors b/out_tensor/model.layers.24.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7622a14213fd81277021ef7d926d70d7e75b9062
--- /dev/null
+++ b/out_tensor/model.layers.24.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c382d2130ff7727e5729accdacf4ab5ee4401e32a502886b5e2559bb0a66d105
+size 2229536
diff --git a/out_tensor/model.layers.25.mlp.down_proj.safetensors b/out_tensor/model.layers.25.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2aea6c4f42ade6e02e34ddc8b2a346700148628c
--- /dev/null
+++ b/out_tensor/model.layers.25.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f4612abd73df586ca5da73b927e75f7a8e37f2a33b93bdd5f791ed6c4385352
+size 30338208
diff --git a/out_tensor/model.layers.25.mlp.gate_proj.safetensors b/out_tensor/model.layers.25.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..66554ed36607abd94b62f3fb766ac8fbe2782b9c
--- /dev/null
+++ b/out_tensor/model.layers.25.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39873d67b8e323405ef48ee588ec4b744cb0e2195576ca882553b61fe98bd84b
+size 29606616
diff --git a/out_tensor/model.layers.25.mlp.up_proj.safetensors b/out_tensor/model.layers.25.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..330ae21faa7346ddaaa8865681fc9de8b63e77af
--- /dev/null
+++ b/out_tensor/model.layers.25.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16e8f617cb1455a6d1bd6a604d9e93db886adc46c1925bfee50a63952f13e392
+size 36946640
diff --git a/out_tensor/model.layers.25.self_attn.k_proj.safetensors b/out_tensor/model.layers.25.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bb2bc12e9437d1a7e6e09be56c100ec0aea2e1da
--- /dev/null
+++ b/out_tensor/model.layers.25.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f9ba413471167f8dbe8f4fe754fcae6de5c02a2131a148ccc4cf3ab80971faf
+size 1688864
diff --git a/out_tensor/model.layers.25.self_attn.o_proj.safetensors b/out_tensor/model.layers.25.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0b77dc4c3d615bf89089ee0c9b025d4068dad99b
--- /dev/null
+++ b/out_tensor/model.layers.25.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f4ea38343cb078649964b0c4739536dcbd8fa2839aceb6d716dc1874980e4e8
+size 8471264
diff --git a/out_tensor/model.layers.25.self_attn.q_proj.safetensors b/out_tensor/model.layers.25.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ed21ccdec85f51d82d1016ce93d0c84f6c9febfd
--- /dev/null
+++ b/out_tensor/model.layers.25.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d95bfc282f169e154f525bf7287cd051990db06c27fdccbca68ec1ac8b09fe13
+size 7423272
diff --git a/out_tensor/model.layers.25.self_attn.v_proj.safetensors b/out_tensor/model.layers.25.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b067ea81d505d9af515e477d306ce5ad73a42169
--- /dev/null
+++ b/out_tensor/model.layers.25.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:753a732291e03ac446f29203cbf42b9fa264f83b0e26105076c401f7d48a1c64
+size 2229536
diff --git a/out_tensor/model.layers.26.mlp.down_proj.safetensors b/out_tensor/model.layers.26.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..86ba541f4608165466c569d7700b27d5fc022757
--- /dev/null
+++ b/out_tensor/model.layers.26.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61a7361213a7fa496d47e0f3872fe5fd811dfe0f1125e0678044ddf857fa32c2
+size 31059104
diff --git a/out_tensor/model.layers.26.mlp.gate_proj.safetensors b/out_tensor/model.layers.26.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8a6ac5265ca5c65ab06eb55723b4ba63fac9375f
--- /dev/null
+++ b/out_tensor/model.layers.26.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f32fcbdec5d17c22305fa7dd08f00cf88c951502bee9800a74a3fad025ccf6c
+size 29606616
diff --git a/out_tensor/model.layers.26.mlp.up_proj.safetensors b/out_tensor/model.layers.26.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..456cbbd477f96a78afb6da719b288a4748128ae8
--- /dev/null
+++ b/out_tensor/model.layers.26.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0de5da4d715f9f974c97eaf492172c9265a8ced2b46bec5343d35b224be4c9ba
+size 36946640
diff --git a/out_tensor/model.layers.26.self_attn.k_proj.safetensors b/out_tensor/model.layers.26.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e75f9609dd300b65bbb0ef1d83174407207d7679
--- /dev/null
+++ b/out_tensor/model.layers.26.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cea309d857a1462135684dbd4f3323e703055a7aca0e553cf2a634779111e72
+size 1688864
diff --git a/out_tensor/model.layers.26.self_attn.o_proj.safetensors b/out_tensor/model.layers.26.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8e15c0ac66130feb665d4ec8314c34f86c8e1663
--- /dev/null
+++ b/out_tensor/model.layers.26.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3b01188731eca89c96f75a2bd2e931a34cf3d2c41a48a9255edd78b1521a82d
+size 8471264
diff --git a/out_tensor/model.layers.26.self_attn.q_proj.safetensors b/out_tensor/model.layers.26.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0df5ccc7f0fbff27948005f056968ffc805ab91e
--- /dev/null
+++ b/out_tensor/model.layers.26.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c2b1683af15fb4f5839bcea76475de299cffaf039edd823cbaf7b13e55ffb64
+size 7423272
diff --git a/out_tensor/model.layers.26.self_attn.v_proj.safetensors b/out_tensor/model.layers.26.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8efd344abcb1f1dede67ec1862a4717a482cd563
--- /dev/null
+++ b/out_tensor/model.layers.26.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb707cbf5d1389bb4e0e65086baf0877ea720ac009cbe7585ada280412b69ad8
+size 2229536
diff --git a/out_tensor/model.layers.27.mlp.down_proj.safetensors b/out_tensor/model.layers.27.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..41826330bac4e41566eca75292c32bda79900507
--- /dev/null
+++ b/out_tensor/model.layers.27.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb7511c1cd99a763c3666ae268e86f9f4bd9ef13dc172c333d06ad1249366a19
+size 31059104
diff --git a/out_tensor/model.layers.27.mlp.gate_proj.safetensors b/out_tensor/model.layers.27.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..efac366eabade26390070d47ff2e9ee938805e6e
--- /dev/null
+++ b/out_tensor/model.layers.27.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:939d2bbd67adfa8de2ffbff1a378e24665b711c543d00715a465401b46614268
+size 29606616
diff --git a/out_tensor/model.layers.27.mlp.up_proj.safetensors b/out_tensor/model.layers.27.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0b55ddba560e91589b1073b8def2ab61fb0e5603
--- /dev/null
+++ b/out_tensor/model.layers.27.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c9342f7ab79d0bbcf39d9122a2b860f5100fc955bf88691613876b1029ff9f47
+size 36946640
diff --git a/out_tensor/model.layers.27.self_attn.k_proj.safetensors b/out_tensor/model.layers.27.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ebe85979b460faf956fba24748a2d3d1346b2b4a
--- /dev/null
+++ b/out_tensor/model.layers.27.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2656dc10adbf40923887d8ab762f73cef43de59a41aa096ee1d3d72d1be459a9
+size 1688864
diff --git a/out_tensor/model.layers.27.self_attn.o_proj.safetensors b/out_tensor/model.layers.27.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d849ef9cb01fd26c015b1127f46000079761285a
--- /dev/null
+++ b/out_tensor/model.layers.27.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a838d86d800ddef3d49df8ce9e9dcf1bb93f584e5c6179cd7da3d37bd48607e
+size 8668456
diff --git a/out_tensor/model.layers.27.self_attn.q_proj.safetensors b/out_tensor/model.layers.27.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..61c586ccd2f429c0586a9ed9353fd5c4a699f058
--- /dev/null
+++ b/out_tensor/model.layers.27.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52e7abf009e6442f066355a901fe5ca97f35abeec7c1263ebc16d1ba3c8c1433
+size 7423272
diff --git a/out_tensor/model.layers.27.self_attn.v_proj.safetensors b/out_tensor/model.layers.27.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..73ffa1741e10a3e865c0a914e1ffea03e209cb84
--- /dev/null
+++ b/out_tensor/model.layers.27.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20c6322ff1f8f602bb9810c46f8878c6651ef5a00c7d2bb72cb06a5bd70ab612
+size 2229536
diff --git a/out_tensor/model.layers.28.mlp.down_proj.safetensors b/out_tensor/model.layers.28.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4f82000eb34cf949436f7a84d573d6d5ec6af029
--- /dev/null
+++ b/out_tensor/model.layers.28.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fdc2b14c35e53fdbc7b207ec911bbf9819bc01594361bf344764fd8cbff87d8
+size 30338208
diff --git a/out_tensor/model.layers.28.mlp.gate_proj.safetensors b/out_tensor/model.layers.28.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b01914f21b8a1b4d7105880561906fbef268b9b6
--- /dev/null
+++ b/out_tensor/model.layers.28.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb2fa4094e6997aa100f30b0e283c096fb013974ed088e588df6d72f8eab484
+size 29606616
diff --git a/out_tensor/model.layers.28.mlp.up_proj.safetensors b/out_tensor/model.layers.28.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8e5f50075250f196652ce2201951af3d06601b59
--- /dev/null
+++ b/out_tensor/model.layers.28.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72e95f41cfdf25c5d5a4d47e1ca436461ce08f47370689c4c68283d0f6cacedf
+size 30295312
diff --git a/out_tensor/model.layers.28.self_attn.k_proj.safetensors b/out_tensor/model.layers.28.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..512231cc88764c6214f2192feb826d36f9f7ff58
--- /dev/null
+++ b/out_tensor/model.layers.28.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c496fae1c03b68d71f69d864d596b2b743c76faa19fdb4a75cd8f1931ee34f5c
+size 1688864
diff --git a/out_tensor/model.layers.28.self_attn.o_proj.safetensors b/out_tensor/model.layers.28.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4d93552ab86d9792c3805c5290b41e9f29c18d27
--- /dev/null
+++ b/out_tensor/model.layers.28.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89257e8f43cafa8defcd53c009d5667385830e48c5b4d761bb375cbc71fe5f55
+size 8668456
diff --git a/out_tensor/model.layers.28.self_attn.q_proj.safetensors b/out_tensor/model.layers.28.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..655355a0603f5ee097472337ca3361c124e4d3cf
--- /dev/null
+++ b/out_tensor/model.layers.28.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc70349acca60f9f24c4892334f9b8ec3da007eb4f1a5c4ee163dce2b6fccd93
+size 6767912
diff --git a/out_tensor/model.layers.28.self_attn.v_proj.safetensors b/out_tensor/model.layers.28.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..93107085e82087dadc22a1f90012108967bd1220
--- /dev/null
+++ b/out_tensor/model.layers.28.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de85eb75da3c59cc141252f512f3c2df09f645e28db0170b081270e1c1afd944
+size 2229536
diff --git a/out_tensor/model.layers.29.mlp.down_proj.safetensors b/out_tensor/model.layers.29.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5071393790f402895c79ce55405e812358611169
--- /dev/null
+++ b/out_tensor/model.layers.29.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0645f112657765c4e8409a50c9664387f4355e7527c3e9ea29c1a16518b2d493
+size 30338208
diff --git a/out_tensor/model.layers.29.mlp.gate_proj.safetensors b/out_tensor/model.layers.29.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b42621e70d50850c0b69569b1371f31fc7f2d0ba
--- /dev/null
+++ b/out_tensor/model.layers.29.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2719700fb77c67d6675e6a26d2e81da59b27746a4671a360aaac68a710b108dc
+size 29606616
diff --git a/out_tensor/model.layers.29.mlp.up_proj.safetensors b/out_tensor/model.layers.29.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b39c8f49f792538614bde7751f7888974290a606
--- /dev/null
+++ b/out_tensor/model.layers.29.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:910ebed2d57e1dff6feb45a1d9c30df8dc0421f5409b558494f7285dd52c4c10
+size 30295312
diff --git a/out_tensor/model.layers.29.self_attn.k_proj.safetensors b/out_tensor/model.layers.29.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..17e7ddbe8dc11a142434a9909a1674313031ce10
--- /dev/null
+++ b/out_tensor/model.layers.29.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0df8de40def5b40830d160c05d944aaa6b593ee4d873d78a2f96b4a2251c1e9a
+size 1688864
diff --git a/out_tensor/model.layers.29.self_attn.o_proj.safetensors b/out_tensor/model.layers.29.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..61061aa00356785514703301702226a3ed37c550
--- /dev/null
+++ b/out_tensor/model.layers.29.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae954566ab1b9a8cbecf8ca1fcea4e56c78ad31a8b202b22dfc5be47a9b53657
+size 8471264
diff --git a/out_tensor/model.layers.29.self_attn.q_proj.safetensors b/out_tensor/model.layers.29.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0b92e58d8afb2d2071336a1f9a1c24e4e861b78e
--- /dev/null
+++ b/out_tensor/model.layers.29.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:202d7ea4119364f2fc8ab97225766c9da895bd4a129d573d49f8e565656a7def
+size 6702376
diff --git a/out_tensor/model.layers.29.self_attn.v_proj.safetensors b/out_tensor/model.layers.29.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..36b6aa628fe69be27b36aa964c0e7054310c08a3
--- /dev/null
+++ b/out_tensor/model.layers.29.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5d8ebd2207f7155ed78e3a5c41a440918bdd1c89eaba2eaa40b113514720b7d
+size 2654944
diff --git a/out_tensor/model.layers.3.mlp.down_proj.safetensors b/out_tensor/model.layers.3.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b19ae56d5df6fa6910491d38a3f4ae5257a7a82a
--- /dev/null
+++ b/out_tensor/model.layers.3.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:348057040e9f4096c4ff6a4dfed7aaaced13ae1af89b70e09d71c649e91a5995
+size 30338200
diff --git a/out_tensor/model.layers.3.mlp.gate_proj.safetensors b/out_tensor/model.layers.3.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..453d0f4ab73895471967a91459be43922f37e91a
--- /dev/null
+++ b/out_tensor/model.layers.3.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eecc45c34f785ff19ef44f01731ff7bc0d827f680a78c6b69772831ea5353c30
+size 30295320
diff --git a/out_tensor/model.layers.3.mlp.up_proj.safetensors b/out_tensor/model.layers.3.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..98be13539b73b409e7a1943ae6a46d9f6123bf28
--- /dev/null
+++ b/out_tensor/model.layers.3.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51b150e4dacea0c06eb02fb27cd9de5f42be232f9ca84b0fc0bf26f044ec0e4e
+size 36946632
diff --git a/out_tensor/model.layers.3.self_attn.k_proj.safetensors b/out_tensor/model.layers.3.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..187f281ab1a039a42e9c8ddb950198a0c8023f20
--- /dev/null
+++ b/out_tensor/model.layers.3.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0515622328ea0712ce879867bfe34389120c2e9749459a875af99cf685083280
+size 1164576
diff --git a/out_tensor/model.layers.3.self_attn.o_proj.safetensors b/out_tensor/model.layers.3.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d38f1834795a5fa43f66c3f25ccf546bab99b9e3
--- /dev/null
+++ b/out_tensor/model.layers.3.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e40739d378d8bf914ae8daf761c526aa252104e3dc0d95c818eda69a519fa456
+size 8471256
diff --git a/out_tensor/model.layers.3.self_attn.q_proj.safetensors b/out_tensor/model.layers.3.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b25bb2af22f1087757e75b3db9fcf5bb0cf10fec
--- /dev/null
+++ b/out_tensor/model.layers.3.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d78c8f5b91a7b8a40b41d21a89cb55da9d3b7fa37c367325f7b1d896dda40fde
+size 4605216
diff --git a/out_tensor/model.layers.3.self_attn.v_proj.safetensors b/out_tensor/model.layers.3.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..20b8b6036bb37b39ce8eef3e0ae80e28d8849383
--- /dev/null
+++ b/out_tensor/model.layers.3.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32bff13a1815668a41f8dd1922f6fed86f9e7f6b7d91c792a51744b63cded099
+size 2130648
diff --git a/out_tensor/model.layers.30.mlp.down_proj.safetensors b/out_tensor/model.layers.30.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9b2cfc07336d167535ea85f06f1e6ed7e1c736d3
--- /dev/null
+++ b/out_tensor/model.layers.30.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c115270418a61cd18d788286482e81439b4851efa54f50a74f5749e0dc6ca7bc
+size 29648064
diff --git a/out_tensor/model.layers.30.mlp.gate_proj.safetensors b/out_tensor/model.layers.30.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..eb1109aeb9e9f3d0d0ee7cb2b72393fda5dbe838
--- /dev/null
+++ b/out_tensor/model.layers.30.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9be003e448396e3d292c384f547c491712a7bd550532317c65ffc4094b00d91b
+size 29606616
diff --git a/out_tensor/model.layers.30.mlp.up_proj.safetensors b/out_tensor/model.layers.30.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7c4eb1316cfb7f0ef489da5ba1f0f5b250cbaf67
--- /dev/null
+++ b/out_tensor/model.layers.30.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40d2acb8a68f69e7b3fd5a2a67e242d8caff20bd8c871b6137a5b12b2f21c0b5
+size 29606608
diff --git a/out_tensor/model.layers.30.self_attn.k_proj.safetensors b/out_tensor/model.layers.30.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..101bc87416db802bd6ad0c333ed811d0e7bc03d2
--- /dev/null
+++ b/out_tensor/model.layers.30.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:246c720619292a53a1849863bfec80dbe5ca94ec5173637fb65ac2139245f643
+size 1688864
diff --git a/out_tensor/model.layers.30.self_attn.o_proj.safetensors b/out_tensor/model.layers.30.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3c950fd3790ca35059e0a952c1f563bcfe8c5f02
--- /dev/null
+++ b/out_tensor/model.layers.30.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b7547a5d2878ca592b62c42302b4d7cb794cacfa260b4c3f4c42a0dab9ba06f
+size 8471264
diff --git a/out_tensor/model.layers.30.self_attn.q_proj.safetensors b/out_tensor/model.layers.30.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c53d9e0cb093089d1ff6efd594794f22b3a81940
--- /dev/null
+++ b/out_tensor/model.layers.30.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46d80a6d4ae25efea895cf94aba79ef38617cbc8bba5b56af909b0112ee5effc
+size 6702376
diff --git a/out_tensor/model.layers.30.self_attn.v_proj.safetensors b/out_tensor/model.layers.30.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ac72495e9ba3674f16fcec32e56378503dfbe3a4
--- /dev/null
+++ b/out_tensor/model.layers.30.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:542e8ee2fd460e1f8a4a9cfdbfed17b6b017712d2ddee3d5c0f06fd072072f19
+size 2229536
diff --git a/out_tensor/model.layers.31.mlp.down_proj.safetensors b/out_tensor/model.layers.31.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f9ee3c04488f1132e0b3536a7c635c94877703df
--- /dev/null
+++ b/out_tensor/model.layers.31.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c82f03b4976ca27674231d96b9c4f7a7799046596e4f20590ac815b66d05d35
+size 23391392
diff --git a/out_tensor/model.layers.31.mlp.gate_proj.safetensors b/out_tensor/model.layers.31.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d55d3b1adb25d762fcdda7af63ae0f79926c4df8
--- /dev/null
+++ b/out_tensor/model.layers.31.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa1b97ede8eebbee961d5df9737c5dad3e4b4e45b23991ed498dc390bb87d881
+size 29606616
diff --git a/out_tensor/model.layers.31.mlp.up_proj.safetensors b/out_tensor/model.layers.31.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..594ed59c621dba47e6b2d1bbdeb4ed6a92a820ec
--- /dev/null
+++ b/out_tensor/model.layers.31.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df5d3b05b2d48ed4560d19a9cb83666941a9bb4534a01f882522d70d53107b04
+size 22955280
diff --git a/out_tensor/model.layers.31.self_attn.k_proj.safetensors b/out_tensor/model.layers.31.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..15cf0342f82d8c16290aa58ceda77de9a365ad73
--- /dev/null
+++ b/out_tensor/model.layers.31.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55bf4113921068aa92026571597e641461b07065a2393e34cbd24a843ed721d0
+size 1688864
diff --git a/out_tensor/model.layers.31.self_attn.o_proj.safetensors b/out_tensor/model.layers.31.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dcfc893d4a2c1a4ea4c7371bd0eb7d78f57c2068
--- /dev/null
+++ b/out_tensor/model.layers.31.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:902e45f66cac287ab3d45ef8348463e462cc4fdf1bd85a0961f41b1a93cd4d10
+size 6571304
diff --git a/out_tensor/model.layers.31.self_attn.q_proj.safetensors b/out_tensor/model.layers.31.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9406da2ccfcf31c6c41805559f8c14066724d64c
--- /dev/null
+++ b/out_tensor/model.layers.31.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83ae4fbdd973c200a7f4dd0792f9bee6ea6de572e5da2a75ce7382cbc1c57918
+size 6702376
diff --git a/out_tensor/model.layers.31.self_attn.v_proj.safetensors b/out_tensor/model.layers.31.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5d3da0f73788fd40d1b296b62c25cc60d0c73713
--- /dev/null
+++ b/out_tensor/model.layers.31.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f49279ec65ebb30f0d2db61a225ee07001093ec6e6218587bb2315793cc466de
+size 2229536
diff --git a/out_tensor/model.layers.4.mlp.down_proj.safetensors b/out_tensor/model.layers.4.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7d5c84262e086a5ec273782cff4f574ff984e379
--- /dev/null
+++ b/out_tensor/model.layers.4.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac85e198e04a0aa17d28ec035b8370c07502fda6dc5d4926fba49526305f6b0a
+size 30338200
diff --git a/out_tensor/model.layers.4.mlp.gate_proj.safetensors b/out_tensor/model.layers.4.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..288e77990eace04ef488a07ededdce2cef9b7551
--- /dev/null
+++ b/out_tensor/model.layers.4.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba891ef5a60ddaa3a73756d561e7bc38d9e13d4287a1b952e1a0cdba4c6be8f0
+size 29606616
diff --git a/out_tensor/model.layers.4.mlp.up_proj.safetensors b/out_tensor/model.layers.4.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..50d25ea3883b4cad38c3ea083f24360c29a289e0
--- /dev/null
+++ b/out_tensor/model.layers.4.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d0eb4fd72506d8ffd87854cd52e82fd047b498bcccced43cd02c6e3ca5da6c4
+size 36946632
diff --git a/out_tensor/model.layers.4.self_attn.k_proj.safetensors b/out_tensor/model.layers.4.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e6e7ff1a642cb8eb13c9b5a584c1ca38a0f047d7
--- /dev/null
+++ b/out_tensor/model.layers.4.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a419c9ec5b2622522b23c2cf0e8ef28d4f16407b399d6e46200373dc3dab8f1d
+size 1262880
diff --git a/out_tensor/model.layers.4.self_attn.o_proj.safetensors b/out_tensor/model.layers.4.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cb28ec87d9d372f69b175384dabfe0ebed5a5d37
--- /dev/null
+++ b/out_tensor/model.layers.4.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba9edcfd6f2acf453957672646c4bc9987e04663ab5f3ceac3bb5aba6d179a83
+size 8471256
diff --git a/out_tensor/model.layers.4.self_attn.q_proj.safetensors b/out_tensor/model.layers.4.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c071670433d7e1673a914258b327c15110e8d6c6
--- /dev/null
+++ b/out_tensor/model.layers.4.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20e0b239c19b5770b7aaf829d4fe8de1bb478837c0de33fa36d942846e62c3c9
+size 4998432
diff --git a/out_tensor/model.layers.4.self_attn.v_proj.safetensors b/out_tensor/model.layers.4.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..55199eb591f727460aa44a704201cafb9a28f093
--- /dev/null
+++ b/out_tensor/model.layers.4.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc2b72421a24d23333ac5966e261dc81825de9115c19acfdeb26436febaff2c7
+size 2130648
diff --git a/out_tensor/model.layers.5.mlp.down_proj.safetensors b/out_tensor/model.layers.5.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..60589fc1a71b14a847a92b99f0159252aed8ea9d
--- /dev/null
+++ b/out_tensor/model.layers.5.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec14ac8ea83fe1389330c2f2104d6573ce8555bdb315d992b728fe75cf2f38a4
+size 30338200
diff --git a/out_tensor/model.layers.5.mlp.gate_proj.safetensors b/out_tensor/model.layers.5.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..807d5e1690cc4709d65219bfe54906dad7a57331
--- /dev/null
+++ b/out_tensor/model.layers.5.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96e362daab05ac0b2127f35c95e33e502e0d5c2bf58e3d3584ebb7d7adc9161a
+size 29606616
diff --git a/out_tensor/model.layers.5.mlp.up_proj.safetensors b/out_tensor/model.layers.5.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..40225ae6de21880ac95e6cc0fbfc569f56964e18
--- /dev/null
+++ b/out_tensor/model.layers.5.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26521e3a71eb83bc50b97a80ba18c84de87697a46504685f88983523b380d9a3
+size 36946632
diff --git a/out_tensor/model.layers.5.self_attn.k_proj.safetensors b/out_tensor/model.layers.5.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0049c7d98f00b512190fea913abb6ae5b8c518d2
--- /dev/null
+++ b/out_tensor/model.layers.5.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b80f30d385dc8fa6ccf03071ff57d5603f3b2fa9b139c9ba42906520c29e297d
+size 1393952
diff --git a/out_tensor/model.layers.5.self_attn.o_proj.safetensors b/out_tensor/model.layers.5.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5bcb7777396ad7130f443847e5f4cbf01fd7b661
--- /dev/null
+++ b/out_tensor/model.layers.5.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:893d07038794c6e0e035c83b502200d26e006cfe34c869c3381e38eff0f8428f
+size 8471256
diff --git a/out_tensor/model.layers.5.self_attn.q_proj.safetensors b/out_tensor/model.layers.5.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ae659b11c096fd56a7e2cbf892319d761524829f
--- /dev/null
+++ b/out_tensor/model.layers.5.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1db7063fcc9e9345354066dc15a5e68fb7b221b674c9c01a373e114a22db5090
+size 5719328
diff --git a/out_tensor/model.layers.5.self_attn.v_proj.safetensors b/out_tensor/model.layers.5.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..05a6b01b15cf0306f2ec00754897943f1790ce09
--- /dev/null
+++ b/out_tensor/model.layers.5.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c784ba1b56502eef2380590af048266575321170a4f52f6b401d043de79aaa31
+size 2180384
diff --git a/out_tensor/model.layers.6.mlp.down_proj.safetensors b/out_tensor/model.layers.6.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..58c0368d77e995f909d0768c7efc60ffbc10b5ef
--- /dev/null
+++ b/out_tensor/model.layers.6.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:488b959f6068ee05ece86c0139e9ac8dc0995c160e2ffa34466698140a6844eb
+size 30338200
diff --git a/out_tensor/model.layers.6.mlp.gate_proj.safetensors b/out_tensor/model.layers.6.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..59e645fa2eae0764fcbcf1260e9c985894511975
--- /dev/null
+++ b/out_tensor/model.layers.6.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1e4d6b98982cf81eeccb07d91dc88599abd38a2e9b09ba6f06af0c3a385db8f
+size 29606616
diff --git a/out_tensor/model.layers.6.mlp.up_proj.safetensors b/out_tensor/model.layers.6.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1556c39714fa93c90fc39a11df2f1d10ccd8edda
--- /dev/null
+++ b/out_tensor/model.layers.6.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09b0266fa800850ecb195e4d363f07c01c8e268a2331641e5fef0cf6defc8cda
+size 30983440
diff --git a/out_tensor/model.layers.6.self_attn.k_proj.safetensors b/out_tensor/model.layers.6.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1668da5f73480d3002977bc901c760051f7711db
--- /dev/null
+++ b/out_tensor/model.layers.6.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bd98b3e1fdd0f28e1643fd5b7db878627d5fdd5f2d0131d6873a4b7a657d856
+size 1262880
diff --git a/out_tensor/model.layers.6.self_attn.o_proj.safetensors b/out_tensor/model.layers.6.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5eaf6026a464672d7c8aff64c28d967fc670efd9
--- /dev/null
+++ b/out_tensor/model.layers.6.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8984579490c13788e3c1b899c68e4979354a8b3d3e0022f2cce591cf4b3aade8
+size 8471256
diff --git a/out_tensor/model.layers.6.self_attn.q_proj.safetensors b/out_tensor/model.layers.6.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..25dc9e70721ffec1fd6851cefe50534659893e72
--- /dev/null
+++ b/out_tensor/model.layers.6.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:717849ad80a94d748d9cd77e0416ac54d821a2b202a7ac905f9b1a59ca400c52
+size 5522720
diff --git a/out_tensor/model.layers.6.self_attn.v_proj.safetensors b/out_tensor/model.layers.6.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6c91a9109e973ad16d17c1e6dc2c473f8c5339d2
--- /dev/null
+++ b/out_tensor/model.layers.6.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96013518fc4d428c70cdac8a16dff167dcd7873e5d9f14b5ff9391f764904bbe
+size 2130648
diff --git a/out_tensor/model.layers.7.mlp.down_proj.safetensors b/out_tensor/model.layers.7.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..62a804066e1acb6f552bcd1fe18c51f640b11209
--- /dev/null
+++ b/out_tensor/model.layers.7.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1202904d6177b3f8f5b3b2a5fe451c7b89402e470ab530148d0bf0acaad3194e
+size 30338200
diff --git a/out_tensor/model.layers.7.mlp.gate_proj.safetensors b/out_tensor/model.layers.7.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5f3e370e0b81724e116a59a79152fa8a33656355
--- /dev/null
+++ b/out_tensor/model.layers.7.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fabcc1a6b318696cd3b2849e3ff5dd658c247e6552a6a8971464b96f113e0414
+size 29606616
diff --git a/out_tensor/model.layers.7.mlp.up_proj.safetensors b/out_tensor/model.layers.7.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c63b2aef6bf9f2c9c732ae05a740f2b177bf5886
--- /dev/null
+++ b/out_tensor/model.layers.7.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ac265164d5f30c6ec3b24202720fb4be813d4a8b2ba34266496682d6d93ec4b
+size 30983440
diff --git a/out_tensor/model.layers.7.self_attn.k_proj.safetensors b/out_tensor/model.layers.7.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a15e8ee8a09206c6fcea587a6891a80784304228
--- /dev/null
+++ b/out_tensor/model.layers.7.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2adb184e2a89141167d43053b782c238aaa717238b1e11ecf3d3e7107bb1b8a
+size 1393952
diff --git a/out_tensor/model.layers.7.self_attn.o_proj.safetensors b/out_tensor/model.layers.7.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..05a35bbfa279057f24b7aa32e18ddd726b242a80
--- /dev/null
+++ b/out_tensor/model.layers.7.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04286cf1907b895645a43749c50a2ef9aa69e79662e72e087cefd37fdcef7d0a
+size 8668448
diff --git a/out_tensor/model.layers.7.self_attn.q_proj.safetensors b/out_tensor/model.layers.7.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..847eec473f4ec6d6bfa8ff0b8d37513eb68f33d1
--- /dev/null
+++ b/out_tensor/model.layers.7.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53281807d3a4e7196d95f54e1f044f3ba935bfff8089dd7bcc6436d132be3b56
+size 5719328
diff --git a/out_tensor/model.layers.7.self_attn.v_proj.safetensors b/out_tensor/model.layers.7.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..559e93182cfdc3a097b2685f292d1f337b1ebfe7
--- /dev/null
+++ b/out_tensor/model.layers.7.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2dad56861aad9fe8ef70970df5afb351dbab1846024031f4d3bdd78bc25e6e4d
+size 2180384
diff --git a/out_tensor/model.layers.8.mlp.down_proj.safetensors b/out_tensor/model.layers.8.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..23fda20b4edee0756ddc7d0eabb9bddb4edd9e9c
--- /dev/null
+++ b/out_tensor/model.layers.8.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2c428c4d68ebb12bc101713def3b82972b93b68e2697e0c1998f30e947510f5
+size 30338200
diff --git a/out_tensor/model.layers.8.mlp.gate_proj.safetensors b/out_tensor/model.layers.8.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..43f2980f7bdbb45969ed643733ff8844bc53cf8e
--- /dev/null
+++ b/out_tensor/model.layers.8.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17d3e00e37b88c69a39a84ad875415271592255ca6d24ee3e9af2776e57f5694
+size 29606616
diff --git a/out_tensor/model.layers.8.mlp.up_proj.safetensors b/out_tensor/model.layers.8.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9410a3f8315379bf6cffb14276e373ff86c36587
--- /dev/null
+++ b/out_tensor/model.layers.8.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcc3b4206ac3025077c807c114946dbca5a3f0e56a729ff9fdda3f221f2d3fa7
+size 30295312
diff --git a/out_tensor/model.layers.8.self_attn.k_proj.safetensors b/out_tensor/model.layers.8.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9e8573d321228d21845b0c1b5e4169514e21115e
--- /dev/null
+++ b/out_tensor/model.layers.8.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c61344b74d5f7183f4ec58b014b229c3f42869956a70f630598f42c6d192a77f
+size 1393952
diff --git a/out_tensor/model.layers.8.self_attn.o_proj.safetensors b/out_tensor/model.layers.8.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3ed068caefbafcec7b64b5b64d7fb53d920f5af2
--- /dev/null
+++ b/out_tensor/model.layers.8.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dc35e869dcb2c2b4b0bf834037c569a6cc6686808b4578ab6f280f85711fc0d
+size 8668448
diff --git a/out_tensor/model.layers.8.self_attn.q_proj.safetensors b/out_tensor/model.layers.8.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2601f805613273148e652f11c4c9c8648d27a2d6
--- /dev/null
+++ b/out_tensor/model.layers.8.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78f9edcadcff65e1ce1f878eee3cc3d5e57fb5968f692373379e8ceefdf69eeb
+size 5719328
diff --git a/out_tensor/model.layers.8.self_attn.v_proj.safetensors b/out_tensor/model.layers.8.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c44d8d153d11edd1d00501c9d0054a30cfdb0a39
--- /dev/null
+++ b/out_tensor/model.layers.8.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92688c4f182450437763cbb758aa73fee08d2a9d768ebf92164bcb9ea491473f
+size 2130648
diff --git a/out_tensor/model.layers.9.mlp.down_proj.safetensors b/out_tensor/model.layers.9.mlp.down_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..685aee2fb1e4be365548751e2ec4a86bdb706cb7
--- /dev/null
+++ b/out_tensor/model.layers.9.mlp.down_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ec91b98446ad6b063aeba9dab6ba99404a0e597a025e1d1ca29285583f970d3
+size 30338200
diff --git a/out_tensor/model.layers.9.mlp.gate_proj.safetensors b/out_tensor/model.layers.9.mlp.gate_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c7f0f025115f4318dce81ba4b04053aa760e728f
--- /dev/null
+++ b/out_tensor/model.layers.9.mlp.gate_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff75811838d2712867c6b90c43388432fdcbcbd8352b1babf0160d50021daa91
+size 29606616
diff --git a/out_tensor/model.layers.9.mlp.up_proj.safetensors b/out_tensor/model.layers.9.mlp.up_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f1ff0cb672a4eaa07f077f91d8166d5943272556
--- /dev/null
+++ b/out_tensor/model.layers.9.mlp.up_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ba89bc7ff81f4cc7991e490c8ff80642b557f538e654ff9a408bc1fe0cdc12c
+size 30295312
diff --git a/out_tensor/model.layers.9.self_attn.k_proj.safetensors b/out_tensor/model.layers.9.self_attn.k_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..198b64026a844974a84ac9b4dee63940fc9b3bc8
--- /dev/null
+++ b/out_tensor/model.layers.9.self_attn.k_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93d9921e29364f334f2f9a9ff51d3d4b2df19b4eb166d356c6af4c5866298b50
+size 1606360
diff --git a/out_tensor/model.layers.9.self_attn.o_proj.safetensors b/out_tensor/model.layers.9.self_attn.o_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..54c33e25838545fb604ec0b637be69094caa9a1a
--- /dev/null
+++ b/out_tensor/model.layers.9.self_attn.o_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8e6bb318e6bc060cbb1c2e77adb9d6454b0aeb4d9e78bc027a30440d797e156
+size 8668448
diff --git a/out_tensor/model.layers.9.self_attn.q_proj.safetensors b/out_tensor/model.layers.9.self_attn.q_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9d92c82187974d2415444b76605e9c6f30336550
--- /dev/null
+++ b/out_tensor/model.layers.9.self_attn.q_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ec03704d0d4bce9755ab40ba33a05660e3b371c9f9ae05a71c46afbedaf1c92
+size 6374104
diff --git a/out_tensor/model.layers.9.self_attn.v_proj.safetensors b/out_tensor/model.layers.9.self_attn.v_proj.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8241a01f26f3e47d89b817e851999d56b7b8efc7
--- /dev/null
+++ b/out_tensor/model.layers.9.self_attn.v_proj.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d929542150ce489d5f337adeda05de3d568adf546750832bca15793b2057d9b9
+size 2130648
diff --git a/pytorch_model-00001-of-00008.bin b/pytorch_model-00001-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..6e791e1e2e1f3cc9e4a3a09bae5ca9b2b147a90f
--- /dev/null
+++ b/pytorch_model-00001-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e6e08a052e25fc9766d3c8dc03f9badf7ed78f5ac8bf667bee18d223464a65d
+size 1889594419
diff --git a/pytorch_model-00002-of-00008.bin b/pytorch_model-00002-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..5a4098ff80fad9dc8c732308c82868a322b6184f
--- /dev/null
+++ b/pytorch_model-00002-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a211772beab275d1696c295a77cc0fd0b9e5e585d004bdb23541bdf1985a4157
+size 1946253333
diff --git a/pytorch_model-00003-of-00008.bin b/pytorch_model-00003-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..73713d3a4326f882ada86214807e3d0839b638c6
--- /dev/null
+++ b/pytorch_model-00003-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:715d009ddd4e40d069fc88904eaa6f029fbf87bd7b2b985367618eb2f99c17d0
+size 1979789691
diff --git a/pytorch_model-00004-of-00008.bin b/pytorch_model-00004-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..359e5cd7d7cfedaf7a2a515ebab1e0345cdcb0cb
--- /dev/null
+++ b/pytorch_model-00004-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6ea860edada7b276bd174f095a03f86fe0c4fdd6d09831d9db2f77fdb3333ec
+size 1946253397
diff --git a/pytorch_model-00005-of-00008.bin b/pytorch_model-00005-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..bb8b499f1a2765e5f18300bc19da26e1baf1abc2
--- /dev/null
+++ b/pytorch_model-00005-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:660240279eed274f7da19d6689621b2a607b05bd136b50d58be93e1401c7856a
+size 1979789691
diff --git a/pytorch_model-00006-of-00008.bin b/pytorch_model-00006-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..3e6299d10a6db04ed1d6c3f4ba63b25f79d61ec2
--- /dev/null
+++ b/pytorch_model-00006-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ee421c50c3494420a11263a4f1d4b8d70bd22dcf21e4f09414efd60cce0b70b
+size 1946253397
diff --git a/pytorch_model-00007-of-00008.bin b/pytorch_model-00007-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..ab4cfe89a7f5b638772575a46df2d4ad56c52195
--- /dev/null
+++ b/pytorch_model-00007-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7145f22d245fdea964607b314efd4a4e7821cc991b7dd728c5154f1121af6c37
+size 1979789691
diff --git a/pytorch_model-00008-of-00008.bin b/pytorch_model-00008-of-00008.bin
new file mode 100644
index 0000000000000000000000000000000000000000..cd8c9ca5c12f00ca761b57fa4d18801dce7f2ee4
--- /dev/null
+++ b/pytorch_model-00008-of-00008.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58ee4fdfec2bf9b297f2943f043461699aed20b2fb1a5632ac168e8866851eb9
+size 815838027
diff --git a/pytorch_model.bin.index.json b/pytorch_model.bin.index.json
new file mode 100644
index 0000000000000000000000000000000000000000..3a5060738fb6bf5c4dda463c55bcbad880710785
--- /dev/null
+++ b/pytorch_model.bin.index.json
@@ -0,0 +1,298 @@
+{
+ "metadata": {
+ "total_size": 14483464192
+ },
+ "weight_map": {
+ "lm_head.weight": "pytorch_model-00008-of-00008.bin",
+ "model.embed_tokens.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.10.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.12.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.mlp.down_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.mlp.up_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.17.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.17.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.17.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00004-of-00008.bin",
+ "model.layers.18.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.20.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.mlp.down_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.21.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.mlp.up_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00005-of-00008.bin",
+ "model.layers.22.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.mlp.down_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.mlp.up_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.26.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.26.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.26.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00006-of-00008.bin",
+ "model.layers.27.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.mlp.down_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.3.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00008.bin",
+ "model.layers.30.input_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.30.mlp.down_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.30.mlp.up_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00007-of-00008.bin",
+ "model.layers.31.input_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.mlp.down_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.mlp.up_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00008-of-00008.bin",
+ "model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.mlp.down_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.mlp.up_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.8.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.8.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.8.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00002-of-00008.bin",
+ "model.layers.9.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.mlp.down_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.mlp.up_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00003-of-00008.bin",
+ "model.norm.weight": "pytorch_model-00008-of-00008.bin"
+ }
+}
diff --git a/special_tokens_map.json b/special_tokens_map.json
new file mode 100644
index 0000000000000000000000000000000000000000..8cd5f1eb30d4e97d74cbf915c36db116aea5eca7
--- /dev/null
+++ b/special_tokens_map.json
@@ -0,0 +1,11 @@
+{
+ "additional_special_tokens": [
+ "",
+ "",
+ ""
+ ],
+ "bos_token": "",
+ "eos_token": "",
+ "pad_token": "",
+ "unk_token": ""
+}
diff --git a/thumbnail.png b/thumbnail.png
new file mode 100644
index 0000000000000000000000000000000000000000..164cdf3cc45c6c502f9f639d3d820b1baff0d0e3
Binary files /dev/null and b/thumbnail.png differ
diff --git a/tokenizer.json b/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..43e6daf936dc0f953cb867ec864adab78f92d9ce
--- /dev/null
+++ b/tokenizer.json
@@ -0,0 +1,91122 @@
+{
+ "version": "1.0",
+ "truncation": null,
+ "padding": null,
+ "added_tokens": [
+ {
+ "id": 0,
+ "content": "",
+ "single_word": false,
+ "lstrip": false,
+ "rstrip": false,
+ "normalized": false,
+ "special": true
+ },
+ {
+ "id": 1,
+ "content": "",
+ "single_word": false,
+ "lstrip": false,
+ "rstrip": false,
+ "normalized": false,
+ "special": true
+ },
+ {
+ "id": 2,
+ "content": "",
+ "single_word": false,
+ "lstrip": false,
+ "rstrip": false,
+ "normalized": false,
+ "special": true
+ }
+ ],
+ "normalizer": {
+ "type": "Sequence",
+ "normalizers": [
+ {
+ "type": "Prepend",
+ "prepend": "▁"
+ },
+ {
+ "type": "Replace",
+ "pattern": {
+ "String": " "
+ },
+ "content": "▁"
+ }
+ ]
+ },
+ "pre_tokenizer": null,
+ "post_processor": {
+ "type": "TemplateProcessing",
+ "single": [
+ {
+ "SpecialToken": {
+ "id": "",
+ "type_id": 0
+ }
+ },
+ {
+ "Sequence": {
+ "id": "A",
+ "type_id": 0
+ }
+ }
+ ],
+ "pair": [
+ {
+ "SpecialToken": {
+ "id": "",
+ "type_id": 0
+ }
+ },
+ {
+ "Sequence": {
+ "id": "A",
+ "type_id": 0
+ }
+ },
+ {
+ "SpecialToken": {
+ "id": "",
+ "type_id": 1
+ }
+ },
+ {
+ "Sequence": {
+ "id": "B",
+ "type_id": 1
+ }
+ }
+ ],
+ "special_tokens": {
+ "": {
+ "id": "",
+ "ids": [
+ 1
+ ],
+ "tokens": [
+ ""
+ ]
+ }
+ }
+ },
+ "decoder": {
+ "type": "Sequence",
+ "decoders": [
+ {
+ "type": "Replace",
+ "pattern": {
+ "String": "▁"
+ },
+ "content": " "
+ },
+ {
+ "type": "ByteFallback"
+ },
+ {
+ "type": "Fuse"
+ },
+ {
+ "type": "Strip",
+ "content": " ",
+ "start": 1,
+ "stop": 0
+ }
+ ]
+ },
+ "model": {
+ "type": "BPE",
+ "dropout": null,
+ "unk_token": "",
+ "continuing_subword_prefix": null,
+ "end_of_word_suffix": null,
+ "fuse_unk": true,
+ "byte_fallback": true,
+ "vocab": {
+ "": 0,
+ "": 1,
+ "": 2,
+ "<0x00>": 3,
+ "<0x01>": 4,
+ "<0x02>": 5,
+ "<0x03>": 6,
+ "<0x04>": 7,
+ "<0x05>": 8,
+ "<0x06>": 9,
+ "<0x07>": 10,
+ "<0x08>": 11,
+ "<0x09>": 12,
+ "<0x0A>": 13,
+ "<0x0B>": 14,
+ "<0x0C>": 15,
+ "<0x0D>": 16,
+ "<0x0E>": 17,
+ "<0x0F>": 18,
+ "<0x10>": 19,
+ "<0x11>": 20,
+ "<0x12>": 21,
+ "<0x13>": 22,
+ "<0x14>": 23,
+ "<0x15>": 24,
+ "<0x16>": 25,
+ "<0x17>": 26,
+ "<0x18>": 27,
+ "<0x19>": 28,
+ "<0x1A>": 29,
+ "<0x1B>": 30,
+ "<0x1C>": 31,
+ "<0x1D>": 32,
+ "<0x1E>": 33,
+ "<0x1F>": 34,
+ "<0x20>": 35,
+ "<0x21>": 36,
+ "<0x22>": 37,
+ "<0x23>": 38,
+ "<0x24>": 39,
+ "<0x25>": 40,
+ "<0x26>": 41,
+ "<0x27>": 42,
+ "<0x28>": 43,
+ "<0x29>": 44,
+ "<0x2A>": 45,
+ "<0x2B>": 46,
+ "<0x2C>": 47,
+ "<0x2D>": 48,
+ "<0x2E>": 49,
+ "<0x2F>": 50,
+ "<0x30>": 51,
+ "<0x31>": 52,
+ "<0x32>": 53,
+ "<0x33>": 54,
+ "<0x34>": 55,
+ "<0x35>": 56,
+ "<0x36>": 57,
+ "<0x37>": 58,
+ "<0x38>": 59,
+ "<0x39>": 60,
+ "<0x3A>": 61,
+ "<0x3B>": 62,
+ "<0x3C>": 63,
+ "<0x3D>": 64,
+ "<0x3E>": 65,
+ "<0x3F>": 66,
+ "<0x40>": 67,
+ "<0x41>": 68,
+ "<0x42>": 69,
+ "<0x43>": 70,
+ "<0x44>": 71,
+ "<0x45>": 72,
+ "<0x46>": 73,
+ "<0x47>": 74,
+ "<0x48>": 75,
+ "<0x49>": 76,
+ "<0x4A>": 77,
+ "<0x4B>": 78,
+ "<0x4C>": 79,
+ "<0x4D>": 80,
+ "<0x4E>": 81,
+ "<0x4F>": 82,
+ "<0x50>": 83,
+ "<0x51>": 84,
+ "<0x52>": 85,
+ "<0x53>": 86,
+ "<0x54>": 87,
+ "<0x55>": 88,
+ "<0x56>": 89,
+ "<0x57>": 90,
+ "<0x58>": 91,
+ "<0x59>": 92,
+ "<0x5A>": 93,
+ "<0x5B>": 94,
+ "<0x5C>": 95,
+ "<0x5D>": 96,
+ "<0x5E>": 97,
+ "<0x5F>": 98,
+ "<0x60>": 99,
+ "<0x61>": 100,
+ "<0x62>": 101,
+ "<0x63>": 102,
+ "<0x64>": 103,
+ "<0x65>": 104,
+ "<0x66>": 105,
+ "<0x67>": 106,
+ "<0x68>": 107,
+ "<0x69>": 108,
+ "<0x6A>": 109,
+ "<0x6B>": 110,
+ "<0x6C>": 111,
+ "<0x6D>": 112,
+ "<0x6E>": 113,
+ "<0x6F>": 114,
+ "<0x70>": 115,
+ "<0x71>": 116,
+ "<0x72>": 117,
+ "<0x73>": 118,
+ "<0x74>": 119,
+ "<0x75>": 120,
+ "<0x76>": 121,
+ "<0x77>": 122,
+ "<0x78>": 123,
+ "<0x79>": 124,
+ "<0x7A>": 125,
+ "<0x7B>": 126,
+ "<0x7C>": 127,
+ "<0x7D>": 128,
+ "<0x7E>": 129,
+ "<0x7F>": 130,
+ "<0x80>": 131,
+ "<0x81>": 132,
+ "<0x82>": 133,
+ "<0x83>": 134,
+ "<0x84>": 135,
+ "<0x85>": 136,
+ "<0x86>": 137,
+ "<0x87>": 138,
+ "<0x88>": 139,
+ "<0x89>": 140,
+ "<0x8A>": 141,
+ "<0x8B>": 142,
+ "<0x8C>": 143,
+ "<0x8D>": 144,
+ "<0x8E>": 145,
+ "<0x8F>": 146,
+ "<0x90>": 147,
+ "<0x91>": 148,
+ "<0x92>": 149,
+ "<0x93>": 150,
+ "<0x94>": 151,
+ "<0x95>": 152,
+ "<0x96>": 153,
+ "<0x97>": 154,
+ "<0x98>": 155,
+ "<0x99>": 156,
+ "<0x9A>": 157,
+ "<0x9B>": 158,
+ "<0x9C>": 159,
+ "<0x9D>": 160,
+ "<0x9E>": 161,
+ "<0x9F>": 162,
+ "<0xA0>": 163,
+ "<0xA1>": 164,
+ "<0xA2>": 165,
+ "<0xA3>": 166,
+ "<0xA4>": 167,
+ "<0xA5>": 168,
+ "<0xA6>": 169,
+ "<0xA7>": 170,
+ "<0xA8>": 171,
+ "<0xA9>": 172,
+ "<0xAA>": 173,
+ "<0xAB>": 174,
+ "<0xAC>": 175,
+ "<0xAD>": 176,
+ "<0xAE>": 177,
+ "<0xAF>": 178,
+ "<0xB0>": 179,
+ "<0xB1>": 180,
+ "<0xB2>": 181,
+ "<0xB3>": 182,
+ "<0xB4>": 183,
+ "<0xB5>": 184,
+ "<0xB6>": 185,
+ "<0xB7>": 186,
+ "<0xB8>": 187,
+ "<0xB9>": 188,
+ "<0xBA>": 189,
+ "<0xBB>": 190,
+ "<0xBC>": 191,
+ "<0xBD>": 192,
+ "<0xBE>": 193,
+ "<0xBF>": 194,
+ "<0xC0>": 195,
+ "<0xC1>": 196,
+ "<0xC2>": 197,
+ "<0xC3>": 198,
+ "<0xC4>": 199,
+ "<0xC5>": 200,
+ "<0xC6>": 201,
+ "<0xC7>": 202,
+ "<0xC8>": 203,
+ "<0xC9>": 204,
+ "<0xCA>": 205,
+ "<0xCB>": 206,
+ "<0xCC>": 207,
+ "<0xCD>": 208,
+ "<0xCE>": 209,
+ "<0xCF>": 210,
+ "<0xD0>": 211,
+ "<0xD1>": 212,
+ "<0xD2>": 213,
+ "<0xD3>": 214,
+ "<0xD4>": 215,
+ "<0xD5>": 216,
+ "<0xD6>": 217,
+ "<0xD7>": 218,
+ "<0xD8>": 219,
+ "<0xD9>": 220,
+ "<0xDA>": 221,
+ "<0xDB>": 222,
+ "<0xDC>": 223,
+ "<0xDD>": 224,
+ "<0xDE>": 225,
+ "<0xDF>": 226,
+ "<0xE0>": 227,
+ "<0xE1>": 228,
+ "<0xE2>": 229,
+ "<0xE3>": 230,
+ "<0xE4>": 231,
+ "<0xE5>": 232,
+ "<0xE6>": 233,
+ "<0xE7>": 234,
+ "<0xE8>": 235,
+ "<0xE9>": 236,
+ "<0xEA>": 237,
+ "<0xEB>": 238,
+ "<0xEC>": 239,
+ "<0xED>": 240,
+ "<0xEE>": 241,
+ "<0xEF>": 242,
+ "<0xF0>": 243,
+ "<0xF1>": 244,
+ "<0xF2>": 245,
+ "<0xF3>": 246,
+ "<0xF4>": 247,
+ "<0xF5>": 248,
+ "<0xF6>": 249,
+ "<0xF7>": 250,
+ "<0xF8>": 251,
+ "<0xF9>": 252,
+ "<0xFA>": 253,
+ "<0xFB>": 254,
+ "<0xFC>": 255,
+ "<0xFD>": 256,
+ "<0xFE>": 257,
+ "<0xFF>": 258,
+ "▁▁": 259,
+ "▁▁▁▁": 260,
+ "▁t": 261,
+ "in": 262,
+ "er": 263,
+ "▁a": 264,
+ "he": 265,
+ "on": 266,
+ "re": 267,
+ "▁s": 268,
+ "en": 269,
+ "at": 270,
+ "or": 271,
+ "▁the": 272,
+ "▁▁▁▁▁▁▁▁": 273,
+ "es": 274,
+ "▁w": 275,
+ "an": 276,
+ "▁c": 277,
+ "is": 278,
+ "it": 279,
+ "ou": 280,
+ "▁d": 281,
+ "al": 282,
+ "ar": 283,
+ "▁p": 284,
+ "▁f": 285,
+ "ed": 286,
+ "▁b": 287,
+ "ing": 288,
+ "▁o": 289,
+ "▁m": 290,
+ "le": 291,
+ "nd": 292,
+ "as": 293,
+ "ic": 294,
+ "▁h": 295,
+ "ion": 296,
+ "▁in": 297,
+ "▁to": 298,
+ "et": 299,
+ "om": 300,
+ "el": 301,
+ "▁of": 302,
+ "st": 303,
+ "▁and": 304,
+ "▁l": 305,
+ "▁th": 306,
+ "▁n": 307,
+ "ent": 308,
+ "il": 309,
+ "ct": 310,
+ "ro": 311,
+ "▁re": 312,
+ "id": 313,
+ "am": 314,
+ "▁I": 315,
+ "ad": 316,
+ "▁e": 317,
+ "▁S": 318,
+ "▁g": 319,
+ "▁T": 320,
+ "im": 321,
+ "ot": 322,
+ "ac": 323,
+ "ur": 324,
+ "▁(": 325,
+ "ig": 326,
+ "▁=": 327,
+ "ol": 328,
+ "ut": 329,
+ "▁A": 330,
+ "se": 331,
+ "▁u": 332,
+ "ve": 333,
+ "▁C": 334,
+ "if": 335,
+ "ow": 336,
+ "▁y": 337,
+ "ch": 338,
+ "ay": 339,
+ "▁de": 340,
+ "▁st": 341,
+ "▁|": 342,
+ "ver": 343,
+ ");": 344,
+ "▁\"": 345,
+ "ly": 346,
+ "▁be": 347,
+ "**": 348,
+ "▁is": 349,
+ "od": 350,
+ "▁M": 351,
+ "ation": 352,
+ "ul": 353,
+ "▁for": 354,
+ "▁▁▁▁▁": 355,
+ "▁on": 356,
+ "ag": 357,
+ "ce": 358,
+ "▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁": 359,
+ "ter": 360,
+ "ir": 361,
+ "th": 362,
+ "▁v": 363,
+ "qu": 364,
+ "▁B": 365,
+ "em": 366,
+ "▁P": 367,
+ "▁you": 368,
+ "▁that": 369,
+ "un": 370,
+ "▁{": 371,
+ "ith": 372,
+ "ri": 373,
+ "est": 374,
+ "ab": 375,
+ "--": 376,
+ "ap": 377,
+ "▁it": 378,
+ "▁con": 379,
+ "ate": 380,
+ "us": 381,
+ "▁H": 382,
+ "um": 383,
+ "▁D": 384,
+ "os": 385,
+ "pe": 386,
+ "▁-": 387,
+ "▁wh": 388,
+ "▁al": 389,
+ "▁as": 390,
+ "and": 391,
+ "ist": 392,
+ "▁L": 393,
+ "▁W": 394,
+ "▁with": 395,
+ "▁an": 396,
+ "ere": 397,
+ "▁*": 398,
+ "▁R": 399,
+ "▁he": 400,
+ "▁F": 401,
+ "oc": 402,
+ "▁was": 403,
+ "ers": 404,
+ "ke": 405,
+ "out": 406,
+ "ht": 407,
+ "▁r": 408,
+ "ess": 409,
+ "op": 410,
+ "res": 411,
+ "ie": 412,
+ "▁E": 413,
+ "▁\\": 414,
+ "▁The": 415,
+ "end": 416,
+ "ld": 417,
+ "▁N": 418,
+ "ort": 419,
+ "▁G": 420,
+ "//": 421,
+ "▁#": 422,
+ "our": 423,
+ "te": 424,
+ "ill": 425,
+ "ain": 426,
+ "▁se": 427,
+ "▁▁▁▁▁▁": 428,
+ "▁$": 429,
+ "▁pro": 430,
+ "ore": 431,
+ "▁com": 432,
+ "ame": 433,
+ "tr": 434,
+ "▁ne": 435,
+ "rom": 436,
+ "ub": 437,
+ "▁at": 438,
+ "▁ex": 439,
+ "ant": 440,
+ "ue": 441,
+ "▁or": 442,
+ "▁}": 443,
+ "art": 444,
+ "ction": 445,
+ "▁k": 446,
+ "pt": 447,
+ "nt": 448,
+ "iv": 449,
+ "de": 450,
+ "▁O": 451,
+ "pl": 452,
+ "urn": 453,
+ "ight": 454,
+ "all": 455,
+ "▁this": 456,
+ "ser": 457,
+ "ave": 458,
+ "▁not": 459,
+ "▁are": 460,
+ "▁j": 461,
+ "▁le": 462,
+ "iz": 463,
+ "▁'": 464,
+ "age": 465,
+ "ment": 466,
+ "▁tr": 467,
+ "ack": 468,
+ "ust": 469,
+ "()": 470,
+ "->": 471,
+ "ity": 472,
+ "ine": 473,
+ "ould": 474,
+ "▁J": 475,
+ "og": 476,
+ "▁from": 477,
+ "▁we": 478,
+ "ell": 479,
+ "▁sh": 480,
+ "▁en": 481,
+ "ure": 482,
+ "port": 483,
+ "▁ch": 484,
+ "ne": 485,
+ "▁by": 486,
+ "per": 487,
+ "ard": 488,
+ "ass": 489,
+ "ge": 490,
+ "ak": 491,
+ "are": 492,
+ "ok": 493,
+ "av": 494,
+ "ive": 495,
+ "ff": 496,
+ "ies": 497,
+ "ath": 498,
+ "turn": 499,
+ "▁U": 500,
+ "int": 501,
+ "----": 502,
+ "▁im": 503,
+ "ost": 504,
+ "ial": 505,
+ "▁have": 506,
+ "ind": 507,
+ "ip": 508,
+ "ans": 509,
+ "xt": 510,
+ "▁do": 511,
+ "cl": 512,
+ "▁if": 513,
+ "con": 514,
+ "ia": 515,
+ "▁his": 516,
+ "ult": 517,
+ "rou": 518,
+ "▁su": 519,
+ "ra": 520,
+ "▁un": 521,
+ "able": 522,
+ "▁<": 523,
+ "▁K": 524,
+ "ome": 525,
+ "▁qu": 526,
+ "get": 527,
+ "▁me": 528,
+ "ast": 529,
+ "ect": 530,
+ "▁##": 531,
+ "to": 532,
+ "▁cl": 533,
+ "▁ab": 534,
+ "ice": 535,
+ "ire": 536,
+ "ber": 537,
+ "one": 538,
+ "ich": 539,
+ "hen": 540,
+ "▁can": 541,
+ "▁Th": 542,
+ "▁la": 543,
+ "▁all": 544,
+ "ime": 545,
+ "ile": 546,
+ "ide": 547,
+ "\",": 548,
+ "▁pl": 549,
+ "▁V": 550,
+ "ru": 551,
+ "orm": 552,
+ "▁had": 553,
+ "ud": 554,
+ "ase": 555,
+ "ord": 556,
+ "),": 557,
+ "▁▁▁▁▁▁▁▁▁▁▁▁": 558,
+ "▁her": 559,
+ "▁In": 560,
+ "ace": 561,
+ "▁but": 562,
+ "ata": 563,
+ "::": 564,
+ "****": 565,
+ "ong": 566,
+ "▁&": 567,
+ "..": 568,
+ "▁▁▁▁▁▁▁▁▁▁▁▁▁": 569,
+ "ite": 570,
+ "ype": 571,
+ "act": 572,
+ "ode": 573,
+ "▁your": 574,
+ "▁out": 575,
+ "▁go": 576,
+ "lic": 577,
+ "ally": 578,
+ "▁so": 579,
+ "ork": 580,
+ "au": 581,
+ "▁up": 582,
+ "▁_": 583,
+ "ll": 584,
+ "==": 585,
+ "▁my": 586,
+ "pp": 587,
+ "cc": 588,
+ "▁//": 589,
+ "▁they": 590,
+ "gh": 591,
+ "▁us": 592,
+ "ib": 593,
+ "ions": 594,
+ "ach": 595,
+ "ens": 596,
+ "▁ar": 597,
+ "ob": 598,
+ "elf": 599,
+ "ook": 600,
+ "ated": 601,
+ "ang": 602,
+ "ign": 603,
+ "▁return": 604,
+ "▁res": 605,
+ "ck": 606,
+ "ous": 607,
+ "ст": 608,
+ ").": 609,
+ "▁п": 610,
+ ".\"": 611,
+ "на": 612,
+ "▁i": 613,
+ "ail": 614,
+ "ep": 615,
+ "▁ad": 616,
+ "ance": 617,
+ "(\"": 618,
+ "▁**": 619,
+ "ther": 620,
+ "ake": 621,
+ "▁will": 622,
+ "▁comp": 623,
+ "▁one": 624,
+ "▁get": 625,
+ "ov": 626,
+ "▁Y": 627,
+ "ary": 628,
+ "ock": 629,
+ "▁she": 630,
+ "che": 631,
+ "ft": 632,
+ "▁new": 633,
+ "▁des": 634,
+ "▁li": 635,
+ "ence": 636,
+ "▁sa": 637,
+ "ress": 638,
+ "▁el": 639,
+ "▁und": 640,
+ "eg": 641,
+ "fer": 642,
+ "ry": 643,
+ "ear": 644,
+ "ose": 645,
+ "very": 646,
+ "',": 647,
+ "▁+": 648,
+ "▁в": 649,
+ "▁He": 650,
+ "ublic": 651,
+ "▁their": 652,
+ "ize": 653,
+ "▁were": 654,
+ "ink": 655,
+ "own": 656,
+ "In": 657,
+ "{\\": 658,
+ "▁has": 659,
+ "▁per": 660,
+ "▁It": 661,
+ "▁St": 662,
+ "her": 663,
+ "ject": 664,
+ "ра": 665,
+ "ild": 666,
+ "so": 667,
+ "▁sp": 668,
+ "ни": 669,
+ "du": 670,
+ "row": 671,
+ "alue": 672,
+ "set": 673,
+ "form": 674,
+ "com": 675,
+ "▁man": 676,
+ "ont": 677,
+ "ull": 678,
+ "▁cont": 679,
+ "▁more": 680,
+ "ick": 681,
+ "▁would": 682,
+ "▁ev": 683,
+ "▁about": 684,
+ "ition": 685,
+ "▁z": 686,
+ "ound": 687,
+ "ree": 688,
+ "▁Ch": 689,
+ "▁which": 690,
+ "io": 691,
+ "();": 692,
+ "▁who": 693,
+ "err": 694,
+ "ory": 695,
+ "ount": 696,
+ "ations": 697,
+ "▁с": 698,
+ "ring": 699,
+ "": 700,
+ "▁fe": 701,
+ "ко": 702,
+ "но": 703,
+ "▁dis": 704,
+ "ma": 705,
+ "▁them": 706,
+ "▁any": 707,
+ "▁no": 708,
+ "--------": 709,
+ "▁pre": 710,
+ "▁te": 711,
+ "▁ro": 712,
+ "▁him": 713,
+ "▁:": 714,
+ "up": 715,
+ "▁int": 716,
+ "▁ag": 717,
+ "St": 718,
+ "ark": 719,
+ "ex": 720,
+ "ph": 721,
+ "ient": 722,
+ "ely": 723,
+ "▁pr": 724,
+ "ER": 725,
+ "▁import": 726,
+ "▁time": 727,
+ "ро": 728,
+ "pro": 729,
+ "User": 730,
+ "lo": 731,
+ "▁/": 732,
+ "▁[": 733,
+ "ors": 734,
+ "=\"": 735,
+ "▁there": 736,
+ "▁like": 737,
+ "old": 738,
+ "▁when": 739,
+ "vers": 740,
+ "▁some": 741,
+ "ings": 742,
+ "))": 743,
+ "▁part": 744,
+ "ical": 745,
+ "▁fun": 746,
+ "▁kn": 747,
+ "ays": 748,
+ "ier": 749,
+ "▁been": 750,
+ "ove": 751,
+ "▁sc": 752,
+ "ian": 753,
+ "▁over": 754,
+ "iel": 755,
+ "▁▁▁▁▁▁▁▁▁▁": 756,
+ "▁pe": 757,
+ "rib": 758,
+ "put": 759,
+ "ec": 760,
+ "eth": 761,
+ "aram": 762,
+ "app": 763,
+ "▁–": 764,
+ "▁stat": 765,
+ "pon": 766,
+ "▁what": 767,
+ "ption": 768,
+ "we": 769,
+ "ade": 770,
+ "▁work": 771,
+ "text": 772,
+ "▁said": 773,
+ "▁###": 774,
+ "IN": 775,
+ "▁just": 776,
+ "irst": 777,
+ "▁into": 778,
+ "▁const": 779,
+ "ource": 780,
+ "tt": 781,
+ "ps": 782,
+ "pr": 783,
+ "erv": 784,
+ "itt": 785,
+ "ug": 786,
+ "_{": 787,
+ "ents": 788,
+ "ish": 789,
+ "ener": 790,
+ "▁inter": 791,
+ "ple": 792,
+ "oll": 793,
+ "mer": 794,
+ "ater": 795,
+ "ool": 796,
+ "ef": 797,
+ "▁public": 798,
+ "▁other": 799,
+ "ре": 800,
+ "▁def": 801,
+ "▁@": 802,
+ "го": 803,
+ "oint": 804,
+ "▁off": 805,
+ "oid": 806,
+ "return": 807,
+ "▁set": 808,
+ "wo": 809,
+ "fter": 810,
+ "sh": 811,
+ "********": 812,
+ "▁our": 813,
+ "riv": 814,
+ "iss": 815,
+ "▁We": 816,
+ "ng": 817,
+ "▁ob": 818,
+ "ss": 819,
+ "gr": 820,
+ "▁than": 821,
+ "pect": 822,
+ "ied": 823,
+ "sc": 824,
+ "iew": 825,
+ "der": 826,
+ "yst": 827,
+ "ev": 828,
+ "▁could": 829,
+ "ann": 830,
+ "enc": 831,
+ "ON": 832,
+ "ix": 833,
+ "anc": 834,
+ "▁also": 835,
+ "reat": 836,
+ "▁am": 837,
+ "▁bec": 838,
+ "▁и": 839,
+ "ual": 840,
+ "pec": 841,
+ "▁.": 842,
+ "▁bl": 843,
+ "lect": 844,
+ "ople": 845,
+ "ys": 846,
+ "▁gr": 847,
+ "ict": 848,
+ "ik": 849,
+ "tring": 850,
+ "▁This": 851,
+ "▁back": 852,
+ "▁о": 853,
+ "▁fin": 854,
+ "atch": 855,
+ "Con": 856,
+ "('": 857,
+ "erm": 858,
+ "▁==": 859,
+ "__": 860,
+ "name": 861,
+ ",\"": 862,
+ "▁did": 863,
+ "ise": 864,
+ "▁only": 865,
+ "ruct": 866,
+ "les": 867,
+ "▁then": 868,
+ "ause": 869,
+ "ва": 870,
+ "▁its": 871,
+ "rit": 872,
+ "▁know": 873,
+ "ield": 874,
+ "▁class": 875,
+ "▁>": 876,
+ "▁em": 877,
+ "▁$\\": 878,
+ "▁year": 879,
+ "wn": 880,
+ "},": 881,
+ "▁del": 882,
+ "ale": 883,
+ "ty": 884,
+ "fig": 885,
+ "sp": 886,
+ "hed": 887,
+ "round": 888,
+ "ew": 889,
+ "▁di": 890,
+ "▁der": 891,
+ "ри": 892,
+ "red": 893,
+ "this": 894,
+ "let": 895,
+ "RE": 896,
+ "ax": 897,
+ "fr": 898,
+ "essage": 899,
+ "ough": 900,
+ "▁comm": 901,
+ "fo": 902,
+ "uch": 903,
+ "oy": 904,
+ "▁people": 905,
+ "ystem": 906,
+ "▁first": 907,
+ "▁function": 908,
+ "ange": 909,
+ "▁how": 910,
+ "▁et": 911,
+ "ah": 912,
+ "▁look": 913,
+ "то": 914,
+ "und": 915,
+ "▁under": 916,
+ "ка": 917,
+ "▁!": 918,
+ "ray": 919,
+ "ST": 920,
+ "ific": 921,
+ "ли": 922,
+ "read": 923,
+ "▁bet": 924,
+ "ious": 925,
+ "arg": 926,
+ "▁need": 927,
+ "math": 928,
+ "▁на": 929,
+ "ert": 930,
+ "▁op": 931,
+ "▁acc": 932,
+ "Pro": 933,
+ "▁est": 934,
+ "▁Un": 935,
+ "▁ent": 936,
+ "▁rec": 937,
+ "▁use": 938,
+ "ен": 939,
+ "▁par": 940,
+ "az": 941,
+ "▁д": 942,
+ "▁Wh": 943,
+ "self": 944,
+ "▁ke": 945,
+ "та": 946,
+ "▁want": 947,
+ "▁end": 948,
+ "▁don": 949,
+ "ek": 950,
+ "ren": 951,
+ "Name": 952,
+ "▁=>": 953,
+ "▁app": 954,
+ "▁que": 955,
+ "igh": 956,
+ "▁bu": 957,
+ "equ": 958,
+ "vel": 959,
+ "▁act": 960,
+ "cre": 961,
+ "AT": 962,
+ "▁var": 963,
+ "cess": 964,
+ "====": 965,
+ "Ex": 966,
+ "▁add": 967,
+ "▁mod": 968,
+ "ung": 969,
+ "▁where": 970,
+ "ning": 971,
+ "▁fl": 972,
+ "als": 973,
+ "tern": 974,
+ "}}": 975,
+ "▁Al": 976,
+ "▁pos": 977,
+ "ank": 978,
+ "▁ap": 979,
+ "eng": 980,
+ "▁“": 981,
+ "ble": 982,
+ "▁reg": 983,
+ "^{": 984,
+ "▁She": 985,
+ "▁*/": 986,
+ "ude": 987,
+ "add": 988,
+ "▁two": 989,
+ "▁col": 990,
+ "▁sm": 991,
+ "air": 992,
+ "▁may": 993,
+ "fore": 994,
+ "▁You": 995,
+ "rough": 996,
+ "▁che": 997,
+ "▁att": 998,
+ "oth": 999,
+ "ла": 1000,
+ "▁co": 1001,
+ "ates": 1002,
+ "▁rem": 1003,
+ "ood": 1004,
+ "Type": 1005,
+ "led": 1006,
+ "ful": 1007,
+ "▁self": 1008,
+ "of": 1009,
+ "▁Ar": 1010,
+ "que": 1011,
+ "▁every": 1012,
+ "ref": 1013,
+ "The": 1014,
+ "▁And": 1015,
+ "▁rel": 1016,
+ "OR": 1017,
+ "Id": 1018,
+ "▁even": 1019,
+ "EN": 1020,
+ "▁hand": 1021,
+ "ait": 1022,
+ "▁should": 1023,
+ "▁after": 1024,
+ "▁dif": 1025,
+ "ght": 1026,
+ "ife": 1027,
+ "ator": 1028,
+ "ash": 1029,
+ "ribut": 1030,
+ "umber": 1031,
+ "▁see": 1032,
+ "ms": 1033,
+ "▁call": 1034,
+ "yn": 1035,
+ "dd": 1036,
+ "▁es": 1037,
+ "▁make": 1038,
+ "other": 1039,
+ "▁—": 1040,
+ "\");": 1041,
+ "str": 1042,
+ "▁long": 1043,
+ "lement": 1044,
+ "▁wor": 1045,
+ "its": 1046,
+ "▁If": 1047,
+ "alse": 1048,
+ "ль": 1049,
+ "ward": 1050,
+ "▁по": 1051,
+ "val": 1052,
+ "ons": 1053,
+ "▁Z": 1054,
+ "▁now": 1055,
+ "data": 1056,
+ "amp": 1057,
+ "ense": 1058,
+ "▁through": 1059,
+ "▁down": 1060,
+ "att": 1061,
+ "▁static": 1062,
+ "ics": 1063,
+ "##": 1064,
+ "pos": 1065,
+ "▁void": 1066,
+ "aw": 1067,
+ "oun": 1068,
+ "▁way": 1069,
+ "ible": 1070,
+ "vent": 1071,
+ "ower": 1072,
+ "▁think": 1073,
+ "ts": 1074,
+ "*/": 1075,
+ "▁again": 1076,
+ "ating": 1077,
+ "те": 1078,
+ "ner": 1079,
+ "▁most": 1080,
+ "line": 1081,
+ "ym": 1082,
+ "▁sub": 1083,
+ "erson": 1084,
+ "▁requ": 1085,
+ "AL": 1086,
+ "AR": 1087,
+ "abel": 1088,
+ "ond": 1089,
+ "));": 1090,
+ "▁Se": 1091,
+ "▁But": 1092,
+ "alk": 1093,
+ "▁An": 1094,
+ "new": 1095,
+ "▁because": 1096,
+ "ger": 1097,
+ "ular": 1098,
+ "roup": 1099,
+ "ta": 1100,
+ "...": 1101,
+ "▁cons": 1102,
+ "▁right": 1103,
+ "▁fr": 1104,
+ "be": 1105,
+ "ily": 1106,
+ "ки": 1107,
+ "▁ph": 1108,
+ "ead": 1109,
+ "?\"": 1110,
+ "▁gu": 1111,
+ "▁else": 1112,
+ "▁som": 1113,
+ "rent": 1114,
+ "co": 1115,
+ "ement": 1116,
+ "▁str": 1117,
+ "ault": 1118,
+ "▁з": 1119,
+ "ло": 1120,
+ "sert": 1121,
+ "var": 1122,
+ "type": 1123,
+ "▁Com": 1124,
+ "ле": 1125,
+ "ins": 1126,
+ "me": 1127,
+ "way": 1128,
+ "ident": 1129,
+ "▁prov": 1130,
+ "▁м": 1131,
+ "▁true": 1132,
+ "▁Pro": 1133,
+ "fl": 1134,
+ "▁sl": 1135,
+ "▁As": 1136,
+ "}\\": 1137,
+ "ID": 1138,
+ "ues": 1139,
+ "▁inst": 1140,
+ "▁name": 1141,
+ "ox": 1142,
+ "▁)": 1143,
+ "li": 1144,
+ "ames": 1145,
+ "Res": 1146,
+ "▁sur": 1147,
+ "param": 1148,
+ "▁start": 1149,
+ "aj": 1150,
+ "SE": 1151,
+ "ask": 1152,
+ "IT": 1153,
+ "String": 1154,
+ "▁ass": 1155,
+ "▁play": 1156,
+ "ting": 1157,
+ "ton": 1158,
+ "▁before": 1159,
+ "▁pol": 1160,
+ "arch": 1161,
+ "▁well": 1162,
+ "Com": 1163,
+ "any": 1164,
+ "olog": 1165,
+ "▁err": 1166,
+ "▁these": 1167,
+ "ars": 1168,
+ "eb": 1169,
+ "▁br": 1170,
+ "▁incl": 1171,
+ "▁hel": 1172,
+ "ern": 1173,
+ "ody": 1174,
+ "во": 1175,
+ "▁ind": 1176,
+ "----------------": 1177,
+ "▁data": 1178,
+ "▁good": 1179,
+ "LE": 1180,
+ "],": 1181,
+ "▁av": 1182,
+ "▁ac": 1183,
+ "ider": 1184,
+ "не": 1185,
+ "▁Q": 1186,
+ "▁min": 1187,
+ "▁much": 1188,
+ "ci": 1189,
+ "els": 1190,
+ "▁cur": 1191,
+ "▁value": 1192,
+ "ery": 1193,
+ "uf": 1194,
+ "▁loc": 1195,
+ "reak": 1196,
+ "ative": 1197,
+ "imes": 1198,
+ "Cl": 1199,
+ "▁,": 1200,
+ "▁ser": 1201,
+ "▁die": 1202,
+ "▁trans": 1203,
+ "▁result": 1204,
+ "ext": 1205,
+ "▁aut": 1206,
+ "land": 1207,
+ "▁&&": 1208,
+ "Ch": 1209,
+ "ten": 1210,
+ "}$": 1211,
+ "▁type": 1212,
+ "cond": 1213,
+ "ices": 1214,
+ "▁very": 1215,
+ "▁own": 1216,
+ "▁fil": 1217,
+ "ities": 1218,
+ "▁produ": 1219,
+ "▁read": 1220,
+ "▁form": 1221,
+ "▁case": 1222,
+ "ather": 1223,
+ "ти": 1224,
+ "да": 1225,
+ "ер": 1226,
+ "Th": 1227,
+ "aut": 1228,
+ "▁spec": 1229,
+ "ij": 1230,
+ "bl": 1231,
+ "ility": 1232,
+ "▁é": 1233,
+ "▁er": 1234,
+ "▁does": 1235,
+ "▁here": 1236,
+ "the": 1237,
+ "ures": 1238,
+ "▁%": 1239,
+ "min": 1240,
+ "▁null": 1241,
+ "rap": 1242,
+ "\")": 1243,
+ "rr": 1244,
+ "List": 1245,
+ "right": 1246,
+ "▁User": 1247,
+ "UL": 1248,
+ "ational": 1249,
+ "▁being": 1250,
+ "AN": 1251,
+ "sk": 1252,
+ "▁car": 1253,
+ "ole": 1254,
+ "▁dist": 1255,
+ "plic": 1256,
+ "ollow": 1257,
+ "▁pres": 1258,
+ "▁such": 1259,
+ "ream": 1260,
+ "ince": 1261,
+ "gan": 1262,
+ "▁For": 1263,
+ "\":": 1264,
+ "son": 1265,
+ "rivate": 1266,
+ "▁years": 1267,
+ "▁serv": 1268,
+ "▁made": 1269,
+ "def": 1270,
+ ";\r": 1271,
+ "▁gl": 1272,
+ "▁bel": 1273,
+ "▁list": 1274,
+ "▁cor": 1275,
+ "▁det": 1276,
+ "ception": 1277,
+ "egin": 1278,
+ "▁б": 1279,
+ "▁char": 1280,
+ "trans": 1281,
+ "▁fam": 1282,
+ "▁!=": 1283,
+ "ouse": 1284,
+ "▁dec": 1285,
+ "ica": 1286,
+ "▁many": 1287,
+ "aking": 1288,
+ "▁à": 1289,
+ "▁sim": 1290,
+ "ages": 1291,
+ "uff": 1292,
+ "ased": 1293,
+ "man": 1294,
+ "▁Sh": 1295,
+ "iet": 1296,
+ "irect": 1297,
+ "▁Re": 1298,
+ "▁differ": 1299,
+ "▁find": 1300,
+ "ethod": 1301,
+ "▁\r": 1302,
+ "ines": 1303,
+ "▁inv": 1304,
+ "▁point": 1305,
+ "▁They": 1306,
+ "▁used": 1307,
+ "ctions": 1308,
+ "▁still": 1309,
+ "ió": 1310,
+ "ined": 1311,
+ "▁while": 1312,
+ "It": 1313,
+ "ember": 1314,
+ "▁say": 1315,
+ "▁help": 1316,
+ "▁cre": 1317,
+ "▁x": 1318,
+ "▁Tr": 1319,
+ "ument": 1320,
+ "▁sk": 1321,
+ "ought": 1322,
+ "ually": 1323,
+ "message": 1324,
+ "▁Con": 1325,
+ "▁mon": 1326,
+ "ared": 1327,
+ "work": 1328,
+ "):": 1329,
+ "ister": 1330,
+ "arn": 1331,
+ "ized": 1332,
+ "Data": 1333,
+ "orn": 1334,
+ "▁head": 1335,
+ "DE": 1336,
+ "▁Le": 1337,
+ "▁person": 1338,
+ "ments": 1339,
+ "ength": 1340,
+ "▁false": 1341,
+ "▁med": 1342,
+ "▁De": 1343,
+ "ache": 1344,
+ "ited": 1345,
+ "▁let": 1346,
+ "▁show": 1347,
+ "▁same": 1348,
+ "uss": 1349,
+ "▁gener": 1350,
+ "▁у": 1351,
+ "cur": 1352,
+ "▁real": 1353,
+ "ced": 1354,
+ "\">": 1355,
+ "struct": 1356,
+ "begin": 1357,
+ "cept": 1358,
+ "▁bo": 1359,
+ "ired": 1360,
+ "▁Fr": 1361,
+ "▁stud": 1362,
+ "dev": 1363,
+ "Ar": 1364,
+ "(\\": 1365,
+ "▁Cl": 1366,
+ "ween": 1367,
+ "▁too": 1368,
+ "▁test": 1369,
+ "▁day": 1370,
+ "oh": 1371,
+ "▁follow": 1372,
+ "ature": 1373,
+ "ze": 1374,
+ "ien": 1375,
+ "reg": 1376,
+ "ces": 1377,
+ "uring": 1378,
+ "amb": 1379,
+ "ina": 1380,
+ "cri": 1381,
+ "▁ed": 1382,
+ "SS": 1383,
+ "uck": 1384,
+ "▁/*": 1385,
+ "CT": 1386,
+ "▁There": 1387,
+ "▁take": 1388,
+ "par": 1389,
+ "ule": 1390,
+ "cal": 1391,
+ "for": 1392,
+ "****************": 1393,
+ "source": 1394,
+ "▁those": 1395,
+ "col": 1396,
+ "▁eff": 1397,
+ "mod": 1398,
+ "cont": 1399,
+ "}{": 1400,
+ "▁around": 1401,
+ "press": 1402,
+ "by": 1403,
+ "▁going": 1404,
+ "ponse": 1405,
+ "▁С": 1406,
+ "▁line": 1407,
+ "date": 1408,
+ "code": 1409,
+ "['": 1410,
+ "▁life": 1411,
+ "ason": 1412,
+ "▁using": 1413,
+ "▁val": 1414,
+ "▁du": 1415,
+ "yp": 1416,
+ "▁▁▁▁▁▁▁▁▁▁▁▁▁▁": 1417,
+ "▁On": 1418,
+ "▁found": 1419,
+ "olut": 1420,
+ "']": 1421,
+ "arent": 1422,
+ "▁string": 1423,
+ "▁met": 1424,
+ "▁wr": 1425,
+ "ush": 1426,
+ "string": 1427,
+ "size": 1428,
+ "▁ver": 1429,
+ "▁each": 1430,
+ "value": 1431,
+ "▁last": 1432,
+ "▁got": 1433,
+ "ven": 1434,
+ "back": 1435,
+ "Set": 1436,
+ "ey": 1437,
+ "rol": 1438,
+ "▁cr": 1439,
+ "thing": 1440,
+ "ret": 1441,
+ "és": 1442,
+ "ism": 1443,
+ "▁between": 1444,
+ "Ob": 1445,
+ "ething": 1446,
+ "mp": 1447,
+ "▁lo": 1448,
+ "ats": 1449,
+ "▁New": 1450,
+ "ви": 1451,
+ "ado": 1452,
+ "dex": 1453,
+ "ди": 1454,
+ "▁pass": 1455,
+ "wh": 1456,
+ "▁den": 1457,
+ "Get": 1458,
+ "apt": 1459,
+ "▁ask": 1460,
+ "▁sup": 1461,
+ "Value": 1462,
+ "ны": 1463,
+ "▁try": 1464,
+ "lation": 1465,
+ "day": 1466,
+ "ness": 1467,
+ "ets": 1468,
+ "▁exper": 1469,
+ "Tr": 1470,
+ "▁Mar": 1471,
+ "serv": 1472,
+ "br": 1473,
+ "▁number": 1474,
+ "inal": 1475,
+ "cent": 1476,
+ "/*": 1477,
+ "not": 1478,
+ "ional": 1479,
+ "▁final": 1480,
+ "')": 1481,
+ "▁run": 1482,
+ "over": 1483,
+ "▁never": 1484,
+ "uc": 1485,
+ "▁high": 1486,
+ "yle": 1487,
+ "▁ins": 1488,
+ "▁best": 1489,
+ "ittle": 1490,
+ "ric": 1491,
+ "▁sign": 1492,
+ "▁dem": 1493,
+ "iness": 1494,
+ "gy": 1495,
+ "▁war": 1496,
+ "ished": 1497,
+ "▁giv": 1498,
+ "key": 1499,
+ "▁X": 1500,
+ "($": 1501,
+ "▁child": 1502,
+ "less": 1503,
+ "ways": 1504,
+ "incl": 1505,
+ "rop": 1506,
+ "raw": 1507,
+ "://": 1508,
+ "▁«": 1509,
+ "no": 1510,
+ "indow": 1511,
+ "fe": 1512,
+ "riend": 1513,
+ "▁les": 1514,
+ "▁los": 1515,
+ "file": 1516,
+ "formation": 1517,
+ "ccess": 1518,
+ "▁В": 1519,
+ "na": 1520,
+ "▁il": 1521,
+ "ision": 1522,
+ "ler": 1523,
+ "▁art": 1524,
+ "Cont": 1525,
+ "▁world": 1526,
+ "▁turn": 1527,
+ "▁really": 1528,
+ "▁Ex": 1529,
+ "ма": 1530,
+ "▁П": 1531,
+ "ters": 1532,
+ "arget": 1533,
+ "Err": 1534,
+ "▁happ": 1535,
+ "time": 1536,
+ "▁So": 1537,
+ "div": 1538,
+ "▁didn": 1539,
+ "ada": 1540,
+ "oot": 1541,
+ "})": 1542,
+ "▁sch": 1543,
+ "▁cle": 1544,
+ "▁something": 1545,
+ "().": 1546,
+ "▁cour": 1547,
+ "ever": 1548,
+ "ants": 1549,
+ "▁?": 1550,
+ "To": 1551,
+ "▁`": 1552,
+ "try": 1553,
+ "ux": 1554,
+ "ais": 1555,
+ "ross": 1556,
+ "hip": 1557,
+ "▁rep": 1558,
+ "label": 1559,
+ "▁both": 1560,
+ "*,": 1561,
+ "ott": 1562,
+ "ми": 1563,
+ "ane": 1564,
+ "▁open": 1565,
+ "ww": 1566,
+ "▁come": 1567,
+ "▁ext": 1568,
+ "rem": 1569,
+ "_{\\": 1570,
+ "▁old": 1571,
+ "ched": 1572,
+ "._": 1573,
+ "ME": 1574,
+ "ify": 1575,
+ "gg": 1576,
+ "Col": 1577,
+ "view": 1578,
+ "▁bus": 1579,
+ "▁must": 1580,
+ "▁different": 1581,
+ "log": 1582,
+ "ists": 1583,
+ "roll": 1584,
+ "ai": 1585,
+ "▁за": 1586,
+ "▁system": 1587,
+ "ivers": 1588,
+ "atus": 1589,
+ "ote": 1590,
+ "med": 1591,
+ "].": 1592,
+ "akes": 1593,
+ "RO": 1594,
+ "▁cent": 1595,
+ "gram": 1596,
+ "▁private": 1597,
+ "▁great": 1598,
+ "\";": 1599,
+ "opy": 1600,
+ "▁feel": 1601,
+ "▁How": 1602,
+ "////": 1603,
+ "IC": 1604,
+ "▁dr": 1605,
+ "ains": 1606,
+ "lock": 1607,
+ "En": 1608,
+ "▁Sch": 1609,
+ "▁mat": 1610,
+ "▁home": 1611,
+ "perty": 1612,
+ "test": 1613,
+ "loc": 1614,
+ "▁wom": 1615,
+ "sw": 1616,
+ "arly": 1617,
+ "▁En": 1618,
+ "▁ко": 1619,
+ "den": 1620,
+ "ста": 1621,
+ "▁а": 1622,
+ "eter": 1623,
+ "▁includ": 1624,
+ "ULL": 1625,
+ "▁mem": 1626,
+ "▁po": 1627,
+ "▁little": 1628,
+ "▁arg": 1629,
+ "▁},": 1630,
+ "include": 1631,
+ "eta": 1632,
+ "▁place": 1633,
+ "idth": 1634,
+ "ustom": 1635,
+ "▁||": 1636,
+ "▁tem": 1637,
+ "ried": 1638,
+ "▁fact": 1639,
+ "ience": 1640,
+ "▁Pl": 1641,
+ "opt": 1642,
+ "ele": 1643,
+ "go": 1644,
+ "AC": 1645,
+ "inter": 1646,
+ "========": 1647,
+ "(),": 1648,
+ "ots": 1649,
+ "ral": 1650,
+ "ique": 1651,
+ "aving": 1652,
+ "ml": 1653,
+ "▁thought": 1654,
+ "frac": 1655,
+ "▁care": 1656,
+ "());": 1657,
+ "▁put": 1658,
+ "▁might": 1659,
+ "▁Amer": 1660,
+ "▁(!": 1661,
+ "ample": 1662,
+ "alth": 1663,
+ "▁few": 1664,
+ "▁state": 1665,
+ "sub": 1666,
+ "▁Or": 1667,
+ "];": 1668,
+ "▁size": 1669,
+ "▁Sp": 1670,
+ "▁without": 1671,
+ "▁poss": 1672,
+ "eq": 1673,
+ "play": 1674,
+ "▁expect": 1675,
+ "▁second": 1676,
+ "▁String": 1677,
+ "uild": 1678,
+ "▁next": 1679,
+ "++": 1680,
+ "requ": 1681,
+ "▁All": 1682,
+ "▁men": 1683,
+ "▁When": 1684,
+ "iter": 1685,
+ "ament": 1686,
+ "net": 1687,
+ "▁К": 1688,
+ "ron": 1689,
+ "aint": 1690,
+ "▁Is": 1691,
+ "ве": 1692,
+ "pend": 1693,
+ "translation": 1694,
+ "▁го": 1695,
+ "че": 1696,
+ "▁van": 1697,
+ "▁another": 1698,
+ "▁ret": 1699,
+ "▁La": 1700,
+ "Mod": 1701,
+ "ION": 1702,
+ "list": 1703,
+ "▁post": 1704,
+ "da": 1705,
+ "ware": 1706,
+ "▁word": 1707,
+ "Error": 1708,
+ "▁seem": 1709,
+ "▁contin": 1710,
+ "atic": 1711,
+ "▁three": 1712,
+ "Object": 1713,
+ "▁partic": 1714,
+ "$.": 1715,
+ "▁mark": 1716,
+ "▁vis": 1717,
+ "rc": 1718,
+ "▁sw": 1719,
+ "ptions": 1720,
+ "▁break": 1721,
+ "▁things": 1722,
+ "ute": 1723,
+ "ui": 1724,
+ "▁That": 1725,
+ "urs": 1726,
+ "gl": 1727,
+ "ру": 1728,
+ "▁file": 1729,
+ "use": 1730,
+ "igned": 1731,
+ "part": 1732,
+ "Un": 1733,
+ "▁equ": 1734,
+ "(&": 1735,
+ "▁lead": 1736,
+ "rm": 1737,
+ "ained": 1738,
+ "▁Be": 1739,
+ "path": 1740,
+ "▁small": 1741,
+ "ager": 1742,
+ "▁always": 1743,
+ "▁El": 1744,
+ "▁order": 1745,
+ "▁ey": 1746,
+ "▁won": 1747,
+ "ape": 1748,
+ "▁left": 1749,
+ "ava": 1750,
+ "item": 1751,
+ "hor": 1752,
+ "▁away": 1753,
+ "bb": 1754,
+ "fun": 1755,
+ "▁Ind": 1756,
+ "mb": 1757,
+ "▁struct": 1758,
+ "▁process": 1759,
+ "▁support": 1760,
+ ");\r": 1761,
+ "ión": 1762,
+ "LO": 1763,
+ "▁oper": 1764,
+ "UT": 1765,
+ "▁·": 1766,
+ "PE": 1767,
+ "load": 1768,
+ "off": 1769,
+ "▁No": 1770,
+ "ives": 1771,
+ "ican": 1772,
+ "▁ve": 1773,
+ "action": 1774,
+ "';": 1775,
+ "▁vo": 1776,
+ "$,": 1777,
+ "▁Gr": 1778,
+ "pre": 1779,
+ "ny": 1780,
+ "aining": 1781,
+ "ior": 1782,
+ "init": 1783,
+ "lection": 1784,
+ "arm": 1785,
+ "umn": 1786,
+ "ags": 1787,
+ "ци": 1788,
+ "ско": 1789,
+ "version": 1790,
+ "▁To": 1791,
+ "▁ref": 1792,
+ "stand": 1793,
+ "▁At": 1794,
+ "ift": 1795,
+ "▁ein": 1796,
+ "face": 1797,
+ "bo": 1798,
+ "ified": 1799,
+ "ved": 1800,
+ "sum": 1801,
+ "une": 1802,
+ "ital": 1803,
+ "ump": 1804,
+ "comm": 1805,
+ "▁mov": 1806,
+ "elt": 1807,
+ "▁von": 1808,
+ "velop": 1809,
+ "ctor": 1810,
+ "head": 1811,
+ "cle": 1812,
+ "▁build": 1813,
+ "inc": 1814,
+ ".'": 1815,
+ "bs": 1816,
+ "info": 1817,
+ "chn": 1818,
+ "▁week": 1819,
+ "▁book": 1820,
+ "HE": 1821,
+ "bar": 1822,
+ "icense": 1823,
+ "▁What": 1824,
+ "▁quest": 1825,
+ "urch": 1826,
+ "ato": 1827,
+ "left": 1828,
+ "▁mar": 1829,
+ "▁top": 1830,
+ "FF": 1831,
+ "▁friend": 1832,
+ "▁beh": 1833,
+ "▁field": 1834,
+ "▁against": 1835,
+ "ract": 1836,
+ "ization": 1837,
+ "user": 1838,
+ "chen": 1839,
+ "▁keep": 1840,
+ "AD": 1841,
+ "itor": 1842,
+ "▁non": 1843,
+ "ird": 1844,
+ "ope": 1845,
+ "▁rest": 1846,
+ "▁dev": 1847,
+ "▁__": 1848,
+ "▁una": 1849,
+ "▁term": 1850,
+ "IS": 1851,
+ "▁pop": 1852,
+ "rist": 1853,
+ "▁since": 1854,
+ "ves": 1855,
+ "▁hard": 1856,
+ "pi": 1857,
+ "util": 1858,
+ "▁soc": 1859,
+ "ene": 1860,
+ "Exception": 1861,
+ "▁local": 1862,
+ "▁direct": 1863,
+ "▁sure": 1864,
+ "▁bro": 1865,
+ "▁da": 1866,
+ "▁": 1867,
+ "▁current": 1868,
+ "':": 1869,
+ "Wh": 1870,
+ "▁information": 1871,
+ "▁ide": 1872,
+ "▁better": 1873,
+ "Text": 1874,
+ "raph": 1875,
+ "▁stand": 1876,
+ "▁check": 1877,
+ "▁к": 1878,
+ "▁na": 1879,
+ "((": 1880,
+ "outh": 1881,
+ "aps": 1882,
+ "▁unt": 1883,
+ "bf": 1884,
+ "▁conf": 1885,
+ "▁spe": 1886,
+ "itle": 1887,
+ "▁Col": 1888,
+ "class": 1889,
+ "ural": 1890,
+ "bers": 1891,
+ "MA": 1892,
+ "ession": 1893,
+ "▁М": 1894,
+ "Info": 1895,
+ "▁Br": 1896,
+ "▁eas": 1897,
+ "ervice": 1898,
+ "aus": 1899,
+ "ari": 1900,
+ "по": 1901,
+ "▁coun": 1902,
+ "де": 1903,
+ "())": 1904,
+ "ling": 1905,
+ "ED": 1906,
+ "ably": 1907,
+ "▁pat": 1908,
+ "org": 1909,
+ "▁id": 1910,
+ "▁г": 1911,
+ "▁tell": 1912,
+ "lex": 1913,
+ "▁allow": 1914,
+ "reen": 1915,
+ "my": 1916,
+ "▁consider": 1917,
+ "▁team": 1918,
+ "lease": 1919,
+ "htt": 1920,
+ "▁Pr": 1921,
+ "/**": 1922,
+ "▁sing": 1923,
+ "Requ": 1924,
+ "Re": 1925,
+ "ides": 1926,
+ "ches": 1927,
+ "▁object": 1928,
+ "ially": 1929,
+ "By": 1930,
+ "ся": 1931,
+ "ided": 1932,
+ "▁free": 1933,
+ "▁proble": 1934,
+ "cite": 1935,
+ "▁);": 1936,
+ "ission": 1937,
+ "▁during": 1938,
+ "▁--": 1939,
+ "ither": 1940,
+ "ля": 1941,
+ "▁leg": 1942,
+ "▁sit": 1943,
+ "ically": 1944,
+ "▁key": 1945,
+ "leg": 1946,
+ "tra": 1947,
+ "▁mom": 1948,
+ "▁expl": 1949,
+ "▁develop": 1950,
+ "▁event": 1951,
+ "▁NULL": 1952,
+ "ohn": 1953,
+ "▁///": 1954,
+ "▁business": 1955,
+ "ча": 1956,
+ "▁prof": 1957,
+ "error": 1958,
+ "▁por": 1959,
+ "▁commun": 1960,
+ "Ind": 1961,
+ "ium": 1962,
+ "Test": 1963,
+ "▁Ad": 1964,
+ "ouble": 1965,
+ "▁son": 1966,
+ "rite": 1967,
+ "ready": 1968,
+ "▁{\r": 1969,
+ "▁thing": 1970,
+ "ня": 1971,
+ "▁Ph": 1972,
+ "ped": 1973,
+ "сь": 1974,
+ "ived": 1975,
+ "You": 1976,
+ "arl": 1977,
+ "const": 1978,
+ "../": 1979,
+ "Se": 1980,
+ "Sh": 1981,
+ "▁power": 1982,
+ "ribute": 1983,
+ "▁My": 1984,
+ "▁talk": 1985,
+ "itch": 1986,
+ "▁called": 1987,
+ "▁came": 1988,
+ "▁belie": 1989,
+ "UR": 1990,
+ "Add": 1991,
+ "▁Res": 1992,
+ "aster": 1993,
+ "ella": 1994,
+ "obal": 1995,
+ "▁until": 1996,
+ "▁hum": 1997,
+ "CO": 1998,
+ "ately": 1999,
+ "####": 2000,
+ "public": 2001,
+ "[]": 2002,
+ "▁room": 2003,
+ "len": 2004,
+ "▁family": 2005,
+ "por": 2006,
+ "▁program": 2007,
+ "▁hist": 2008,
+ "▁mus": 2009,
+ "arge": 2010,
+ "oney": 2011,
+ "Im": 2012,
+ "else": 2013,
+ "ails": 2014,
+ "af": 2015,
+ "▁love": 2016,
+ "är": 2017,
+ "ases": 2018,
+ "pha": 2019,
+ "ours": 2020,
+ "dis": 2021,
+ "map": 2022,
+ "iver": 2023,
+ "ör": 2024,
+ "▁Bl": 2025,
+ "ateg": 2026,
+ "state": 2027,
+ "State": 2028,
+ "ertain": 2029,
+ "▁effect": 2030,
+ "print": 2031,
+ "▁big": 2032,
+ "index": 2033,
+ "▁pub": 2034,
+ "vert": 2035,
+ "ero": 2036,
+ "md": 2037,
+ "▁method": 2038,
+ "▁game": 2039,
+ "ries": 2040,
+ "lete": 2041,
+ "Item": 2042,
+ "ING": 2043,
+ "resent": 2044,
+ "ality": 2045,
+ "pty": 2046,
+ "ley": 2047,
+ "ocument": 2048,
+ "▁beg": 2049,
+ "TR": 2050,
+ "}.": 2051,
+ "▁school": 2052,
+ "hes": 2053,
+ "до": 2054,
+ "▁lot": 2055,
+ "▁took": 2056,
+ "▁adv": 2057,
+ "▁cap": 2058,
+ "MP": 2059,
+ "unk": 2060,
+ "▁light": 2061,
+ "▁later": 2062,
+ ".,": 2063,
+ "Key": 2064,
+ "itions": 2065,
+ "▁enough": 2066,
+ "▁/**": 2067,
+ "▁went": 2068,
+ "ão": 2069,
+ "▁though": 2070,
+ "▁group": 2071,
+ "▁mean": 2072,
+ "ски": 2073,
+ "AP": 2074,
+ "▁num": 2075,
+ "▁cond": 2076,
+ "ні": 2077,
+ "▁given": 2078,
+ "▁why": 2079,
+ "▁rece": 2080,
+ "▁side": 2081,
+ "▁far": 2082,
+ "Context": 2083,
+ "ме": 2084,
+ "▁log": 2085,
+ "View": 2086,
+ "▁<<": 2087,
+ "fil": 2088,
+ "aces": 2089,
+ "ency": 2090,
+ "oad": 2091,
+ "ered": 2092,
+ "▁product": 2093,
+ "ET": 2094,
+ "▁param": 2095,
+ "▁prote": 2096,
+ "tes": 2097,
+ "Time": 2098,
+ "je": 2099,
+ "olution": 2100,
+ "▁ра": 2101,
+ "▁month": 2102,
+ "ference": 2103,
+ "▁appe": 2104,
+ "▁face": 2105,
+ "ened": 2106,
+ "tract": 2107,
+ "▁less": 2108,
+ "AS": 2109,
+ "ée": 2110,
+ "▁give": 2111,
+ "▁kind": 2112,
+ "▁count": 2113,
+ "count": 2114,
+ "▁stop": 2115,
+ "▁gover": 2116,
+ "ka": 2117,
+ "▁error": 2118,
+ "ences": 2119,
+ "▁mil": 2120,
+ "alf": 2121,
+ "ync": 2122,
+ "vious": 2123,
+ "ho": 2124,
+ "▁night": 2125,
+ "era": 2126,
+ "▁про": 2127,
+ "▁sol": 2128,
+ "men": 2129,
+ "▁water": 2130,
+ "ering": 2131,
+ "▁lim": 2132,
+ "Param": 2133,
+ "▁house": 2134,
+ "▁System": 2135,
+ "▁pay": 2136,
+ "▁:=": 2137,
+ "uro": 2138,
+ "oci": 2139,
+ "zy": 2140,
+ "▁already": 2141,
+ ",\\": 2142,
+ "length": 2143,
+ "▁si": 2144,
+ "▁interest": 2145,
+ "aff": 2146,
+ "cted": 2147,
+ "ention": 2148,
+ "▁до": 2149,
+ "ume": 2150,
+ "▁appro": 2151,
+ "bre": 2152,
+ "IG": 2153,
+ "▁throw": 2154,
+ "mathcal": 2155,
+ "irl": 2156,
+ "▁prom": 2157,
+ "oss": 2158,
+ "▁request": 2159,
+ "equation": 2160,
+ "ology": 2161,
+ "mit": 2162,
+ "▁pack": 2163,
+ "ino": 2164,
+ "array": 2165,
+ "za": 2166,
+ "til": 2167,
+ "UN": 2168,
+ "▁present": 2169,
+ "▁organ": 2170,
+ "File": 2171,
+ "▁orig": 2172,
+ "▁full": 2173,
+ "istr": 2174,
+ "▁flo": 2175,
+ "hr": 2176,
+ "▁assert": 2177,
+ "ards": 2178,
+ "url": 2179,
+ "enn": 2180,
+ "sl": 2181,
+ "▁А": 2182,
+ "▁cho": 2183,
+ "▁level": 2184,
+ "OT": 2185,
+ "word": 2186,
+ "▁body": 2187,
+ "▁user": 2188,
+ "ía": 2189,
+ "Qu": 2190,
+ "▁main": 2191,
+ "AB": 2192,
+ "ploy": 2193,
+ "Event": 2194,
+ "▁super": 2195,
+ "oken": 2196,
+ "▁Н": 2197,
+ "As": 2198,
+ "thers": 2199,
+ "мо": 2200,
+ "ку": 2201,
+ "▁days": 2202,
+ "▁done": 2203,
+ "▁view": 2204,
+ "side": 2205,
+ "си": 2206,
+ "');": 2207,
+ "▁vol": 2208,
+ "▁tot": 2209,
+ "case": 2210,
+ "▁aff": 2211,
+ "Request": 2212,
+ "▁Man": 2213,
+ "\\\\": 2214,
+ "▁John": 2215,
+ "▁Б": 2216,
+ "orth": 2217,
+ "▁je": 2218,
+ "▁une": 2219,
+ "la": 2220,
+ "[\"": 2221,
+ "field": 2222,
+ "▁US": 2223,
+ "ico": 2224,
+ "▁perform": 2225,
+ "ailable": 2226,
+ "Config": 2227,
+ "Or": 2228,
+ "▁model": 2229,
+ "ales": 2230,
+ "▁create": 2231,
+ "▁ann": 2232,
+ "ances": 2233,
+ "IL": 2234,
+ "ination": 2235,
+ "▁Im": 2236,
+ "ante": 2237,
+ "ana": 2238,
+ "ан": 2239,
+ "▁told": 2240,
+ "config": 2241,
+ "\"]": 2242,
+ "met": 2243,
+ "lt": 2244,
+ "▁text": 2245,
+ "▁May": 2246,
+ "▁org": 2247,
+ "▁port": 2248,
+ "Pl": 2249,
+ "ently": 2250,
+ "▁door": 2251,
+ "US": 2252,
+ "▁(*": 2253,
+ "kt": 2254,
+ "ES": 2255,
+ "ential": 2256,
+ "▁iss": 2257,
+ "▁inc": 2258,
+ "Node": 2259,
+ "ively": 2260,
+ "▁asked": 2261,
+ "irt": 2262,
+ "▁Te": 2263,
+ "▁report": 2264,
+ "▁chang": 2265,
+ "сти": 2266,
+ "▁along": 2267,
+ "▁change": 2268,
+ "Size": 2269,
+ "▁ever": 2270,
+ "▁occ": 2271,
+ "ury": 2272,
+ "▁mind": 2273,
+ "order": 2274,
+ "point": 2275,
+ "сто": 2276,
+ "▁whe": 2277,
+ "▁important": 2278,
+ "des": 2279,
+ "▁Not": 2280,
+ "▁writ": 2281,
+ "▁eyes": 2282,
+ "▁desc": 2283,
+ "most": 2284,
+ "ks": 2285,
+ "▁bit": 2286,
+ "▁▁▁": 2287,
+ "▁success": 2288,
+ "ть": 2289,
+ "бо": 2290,
+ "core": 2291,
+ "}(": 2292,
+ "▁array": 2293,
+ "lin": 2294,
+ "lish": 2295,
+ "▁following": 2296,
+ "Field": 2297,
+ "ids": 2298,
+ "hing": 2299,
+ "▁cal": 2300,
+ "Is": 2301,
+ "aring": 2302,
+ "lev": 2303,
+ "alt": 2304,
+ "CH": 2305,
+ "▁dé": 2306,
+ "alpha": 2307,
+ "▁four": 2308,
+ "▁law": 2309,
+ "▁се": 2310,
+ "iron": 2311,
+ "▁disc": 2312,
+ "се": 2313,
+ "ken": 2314,
+ "node": 2315,
+ "▁Par": 2316,
+ "▁Eng": 2317,
+ "▁move": 2318,
+ "▁License": 2319,
+ "cul": 2320,
+ "ione": 2321,
+ ")$": 2322,
+ "▁tw": 2323,
+ "We": 2324,
+ "sel": 2325,
+ "▁With": 2326,
+ "▁once": 2327,
+ "Service": 2328,
+ "bol": 2329,
+ "ured": 2330,
+ "ida": 2331,
+ "▁Qu": 2332,
+ "▁grow": 2333,
+ "▁conne": 2334,
+ "EX": 2335,
+ "▁htt": 2336,
+ "▁};": 2337,
+ "▁walk": 2338,
+ "▁init": 2339,
+ "nal": 2340,
+ "ender": 2341,
+ "cription": 2342,
+ "mber": 2343,
+ "lected": 2344,
+ "po": 2345,
+ "▁nil": 2346,
+ "▁prob": 2347,
+ "чи": 2348,
+ "▁Ste": 2349,
+ "ison": 2350,
+ "ands": 2351,
+ "osed": 2352,
+ "же": 2353,
+ "▁His": 2354,
+ "ür": 2355,
+ "Man": 2356,
+ "Element": 2357,
+ "▁able": 2358,
+ "Index": 2359,
+ "search": 2360,
+ "▁mag": 2361,
+ "ар": 2362,
+ "▁course": 2363,
+ "▁Car": 2364,
+ "▁exp": 2365,
+ "aph": 2366,
+ "▁mit": 2367,
+ "▁doesn": 2368,
+ "▁default": 2369,
+ "/>": 2370,
+ "aim": 2371,
+ "▁service": 2372,
+ "▁within": 2373,
+ "angu": 2374,
+ "▁Д": 2375,
+ "uffer": 2376,
+ "AG": 2377,
+ "▁Do": 2378,
+ "▁incre": 2379,
+ "▁understand": 2380,
+ "}^": 2381,
+ "▁looked": 2382,
+ "gen": 2383,
+ "ailed": 2384,
+ "▁е": 2385,
+ "ayer": 2386,
+ "▁One": 2387,
+ "▁bas": 2388,
+ "▁job": 2389,
+ "mu": 2390,
+ "but": 2391,
+ "elta": 2392,
+ "▁Christ": 2393,
+ "uration": 2394,
+ "▁record": 2395,
+ "▁Univers": 2396,
+ "ivid": 2397,
+ "valid": 2398,
+ "▁Р": 2399,
+ "▁hold": 2400,
+ "▁table": 2401,
+ "ones": 2402,
+ "link": 2403,
+ "▁Ge": 2404,
+ "▁offer": 2405,
+ "ster": 2406,
+ "Form": 2407,
+ "={": 2408,
+ "▁не": 2409,
+ "stance": 2410,
+ "▁govern": 2411,
+ "▁techn": 2412,
+ "▁prim": 2413,
+ "*.": 2414,
+ "cho": 2415,
+ "max": 2416,
+ "▁fore": 2417,
+ "▁Can": 2418,
+ "▁polit": 2419,
+ "ories": 2420,
+ "▁times": 2421,
+ "▁dans": 2422,
+ "▁air": 2423,
+ "▁anything": 2424,
+ "▁sever": 2425,
+ "acy": 2426,
+ "}_": 2427,
+ "He": 2428,
+ "▁least": 2429,
+ "ips": 2430,
+ "ENT": 2431,
+ "do": 2432,
+ "▁от": 2433,
+ "▁cost": 2434,
+ ".”": 2435,
+ "▁children": 2436,
+ "ability": 2437,
+ "But": 2438,
+ "▁path": 2439,
+ "result": 2440,
+ "acter": 2441,
+ "▁element": 2442,
+ "ee": 2443,
+ "▁wait": 2444,
+ "▁money": 2445,
+ "Map": 2446,
+ "td": 2447,
+ "oin": 2448,
+ "iving": 2449,
+ "icht": 2450,
+ "icy": 2451,
+ "sch": 2452,
+ "ste": 2453,
+ "ду": 2454,
+ "ored": 2455,
+ "oud": 2456,
+ "ille": 2457,
+ "ised": 2458,
+ "plication": 2459,
+ "▁custom": 2460,
+ "▁having": 2461,
+ "ponent": 2462,
+ "▁By": 2463,
+ "ules": 2464,
+ "ued": 2465,
+ "atter": 2466,
+ "And": 2467,
+ "itive": 2468,
+ "Def": 2469,
+ "▁moment": 2470,
+ "aterial": 2471,
+ "Class": 2472,
+ "ograph": 2473,
+ "ike": 2474,
+ "▁large": 2475,
+ "▁####": 2476,
+ "▁either": 2477,
+ "duct": 2478,
+ "▁Then": 2479,
+ "▁Gu": 2480,
+ "olean": 2481,
+ "pert": 2482,
+ "▁Get": 2483,
+ "▁Ab": 2484,
+ "▁short": 2485,
+ "On": 2486,
+ "iment": 2487,
+ "▁project": 2488,
+ "cript": 2489,
+ "▁including": 2490,
+ "ния": 2491,
+ "▁making": 2492,
+ "▁someone": 2493,
+ "▁Fl": 2494,
+ "▁sat": 2495,
+ "▁company": 2496,
+ "ocus": 2497,
+ "pu": 2498,
+ "▁God": 2499,
+ "ification": 2500,
+ "No": 2501,
+ "▁sn": 2502,
+ "ano": 2503,
+ "ga": 2504,
+ "▁au": 2505,
+ "▁cou": 2506,
+ "ás": 2507,
+ "ended": 2508,
+ "ту": 2509,
+ "ober": 2510,
+ "▁nothing": 2511,
+ "▁net": 2512,
+ "▁pot": 2513,
+ "▁typ": 2514,
+ "▁item": 2515,
+ "rew": 2516,
+ "Att": 2517,
+ "▁young": 2518,
+ "}\r": 2519,
+ "nder": 2520,
+ "start": 2521,
+ "▁Sc": 2522,
+ "*)": 2523,
+ "▁enc": 2524,
+ "▁women": 2525,
+ "▁looking": 2526,
+ "▁ро": 2527,
+ "▁health": 2528,
+ "Path": 2529,
+ "▁After": 2530,
+ "▁mult": 2531,
+ "▁{\\": 2532,
+ "▁land": 2533,
+ "orld": 2534,
+ "▁Des": 2535,
+ "▁eng": 2536,
+ "input": 2537,
+ "▁Pol": 2538,
+ "\"\"": 2539,
+ "Code": 2540,
+ "▁supp": 2541,
+ "ainer": 2542,
+ "heck": 2543,
+ "▁mor": 2544,
+ "▁mill": 2545,
+ "▁aw": 2546,
+ "fs": 2547,
+ "▁doing": 2548,
+ "tings": 2549,
+ "ades": 2550,
+ "▁toget": 2551,
+ "▁certain": 2552,
+ "▁together": 2553,
+ "CE": 2554,
+ "ideo": 2555,
+ "▁American": 2556,
+ "ony": 2557,
+ "idd": 2558,
+ "II": 2559,
+ "ged": 2560,
+ "ables": 2561,
+ "▁ident": 2562,
+ "iod": 2563,
+ "▁parent": 2564,
+ "For": 2565,
+ "ambda": 2566,
+ "ando": 2567,
+ "=\\": 2568,
+ "aged": 2569,
+ "ending": 2570,
+ "Int": 2571,
+ "▁possible": 2572,
+ "▁со": 2573,
+ "ivity": 2574,
+ "num": 2575,
+ "rt": 2576,
+ "ajor": 2577,
+ "create": 2578,
+ "ride": 2579,
+ "▁knew": 2580,
+ "bit": 2581,
+ "itional": 2582,
+ "▁lik": 2583,
+ "▁Her": 2584,
+ "ension": 2585,
+ "\".": 2586,
+ "oto": 2587,
+ "▁exist": 2588,
+ "aken": 2589,
+ "▁actually": 2590,
+ "ca": 2591,
+ "▁Г": 2592,
+ "хо": 2593,
+ "inn": 2594,
+ "All": 2595,
+ "buf": 2596,
+ "▁Me": 2597,
+ "▁seen": 2598,
+ "ops": 2599,
+ "▁▁▁▁▁▁▁▁▁": 2600,
+ "Not": 2601,
+ "▁control": 2602,
+ "▁respon": 2603,
+ "};": 2604,
+ "ilt": 2605,
+ "isk": 2606,
+ "▁bad": 2607,
+ "▁often": 2608,
+ "▁past": 2609,
+ "aper": 2610,
+ "▁reason": 2611,
+ "eters": 2612,
+ "▁wanted": 2613,
+ "ura": 2614,
+ "table": 2615,
+ "ormal": 2616,
+ "width": 2617,
+ "га": 2618,
+ "ptr": 2619,
+ "▁dest": 2620,
+ "▁design": 2621,
+ "▁sound": 2622,
+ "▁plan": 2623,
+ "▁base": 2624,
+ "hand": 2625,
+ "gs": 2626,
+ "▁says": 2627,
+ "function": 2628,
+ "▁tri": 2629,
+ "mt": 2630,
+ "▁invest": 2631,
+ "▁available": 2632,
+ "ayout": 2633,
+ "▁och": 2634,
+ "▁las": 2635,
+ "illed": 2636,
+ "Val": 2637,
+ "▁ф": 2638,
+ "iety": 2639,
+ "mon": 2640,
+ "Hand": 2641,
+ "Fr": 2642,
+ "iam": 2643,
+ "pace": 2644,
+ "▁Ob": 2645,
+ "▁para": 2646,
+ "▁meet": 2647,
+ "▁sum": 2648,
+ "Message": 2649,
+ "ici": 2650,
+ "▁known": 2651,
+ "▁gen": 2652,
+ "amma": 2653,
+ "arr": 2654,
+ "▁tre": 2655,
+ "oke": 2656,
+ "uth": 2657,
+ "~\\": 2658,
+ "▁experience": 2659,
+ "icle": 2660,
+ "▁Il": 2661,
+ "▁sent": 2662,
+ "▁others": 2663,
+ "▁soft": 2664,
+ "IP": 2665,
+ "▁max": 2666,
+ "ball": 2667,
+ "▁market": 2668,
+ "▁pour": 2669,
+ "pression": 2670,
+ "eps": 2671,
+ "▁saw": 2672,
+ "▁across": 2673,
+ "▁Su": 2674,
+ "Over": 2675,
+ "ние": 2676,
+ "ulation": 2677,
+ "▁Reg": 2678,
+ "▁+=": 2679,
+ "body": 2680,
+ ")\\": 2681,
+ "▁print": 2682,
+ "▁при": 2683,
+ "db": 2684,
+ "ources": 2685,
+ "wards": 2686,
+ "▁black": 2687,
+ "со": 2688,
+ "ili": 2689,
+ "▁Ed": 2690,
+ "▁complet": 2691,
+ "▁single": 2692,
+ "▁IN": 2693,
+ "ached": 2694,
+ "bt": 2695,
+ "▁code": 2696,
+ "▁bool": 2697,
+ "▁area": 2698,
+ "▁require": 2699,
+ "▁problem": 2700,
+ "aced": 2701,
+ "Equ": 2702,
+ "▁config": 2703,
+ "vec": 2704,
+ "ney": 2705,
+ "cy": 2706,
+ "Al": 2707,
+ "▁account": 2708,
+ "ymbol": 2709,
+ "▁ste": 2710,
+ "ges": 2711,
+ "Array": 2712,
+ "empl": 2713,
+ "context": 2714,
+ "Des": 2715,
+ "Result": 2716,
+ "ecut": 2717,
+ "▁target": 2718,
+ "▁getting": 2719,
+ "\"/>": 2720,
+ "ogle": 2721,
+ "▁himself": 2722,
+ "▁wasn": 2723,
+ "▁block": 2724,
+ "▁ant": 2725,
+ "▁York": 2726,
+ "▁become": 2727,
+ "iff": 2728,
+ "ports": 2729,
+ "reate": 2730,
+ "='": 2731,
+ "cd": 2732,
+ "location": 2733,
+ "ет": 2734,
+ "▁access": 2735,
+ "gress": 2736,
+ "ros": 2737,
+ "Up": 2738,
+ "▁working": 2739,
+ "▁Am": 2740,
+ "iqu": 2741,
+ "cer": 2742,
+ "▁((": 2743,
+ "▁Per": 2744,
+ "▁func": 2745,
+ "▁girl": 2746,
+ "▁above": 2747,
+ "pen": 2748,
+ "пи": 2749,
+ "ido": 2750,
+ "▁version": 2751,
+ "TY": 2752,
+ "▁;": 2753,
+ "mary": 2754,
+ "abled": 2755,
+ "annel": 2756,
+ "▁example": 2757,
+ "▁context": 2758,
+ "OP": 2759,
+ "▁red": 2760,
+ "▁cir": 2761,
+ "sm": 2762,
+ "Log": 2763,
+ "▁space": 2764,
+ "▁fut": 2765,
+ "▁Gener": 2766,
+ "ills": 2767,
+ "▁dri": 2768,
+ "_.": 2769,
+ "▁felt": 2770,
+ "▁offic": 2771,
+ "▁===": 2772,
+ "ii": 2773,
+ "▁started": 2774,
+ "▁Т": 2775,
+ "▁});": 2776,
+ "js": 2777,
+ "▁front": 2778,
+ "▁almost": 2779,
+ "irm": 2780,
+ "!\"": 2781,
+ "signed": 2782,
+ "▁yet": 2783,
+ "▁trad": 2784,
+ "ients": 2785,
+ "ama": 2786,
+ "▁input": 2787,
+ "lim": 2788,
+ "па": 2789,
+ "▁ка": 2790,
+ "▁camp": 2791,
+ "ibr": 2792,
+ "fect": 2793,
+ "unt": 2794,
+ "▁half": 2795,
+ "▁cover": 2796,
+ "anguage": 2797,
+ "▁ben": 2798,
+ "ha": 2799,
+ "▁diff": 2800,
+ "_\\": 2801,
+ "▁об": 2802,
+ "])": 2803,
+ "odes": 2804,
+ "hel": 2805,
+ "ios": 2806,
+ "▁О": 2807,
+ "▁mot": 2808,
+ "▁social": 2809,
+ "////////": 2810,
+ "▁stre": 2811,
+ "ground": 2812,
+ "ів": 2813,
+ "object": 2814,
+ "ples": 2815,
+ "reed": 2816,
+ "▁een": 2817,
+ "▁based": 2818,
+ "▁range": 2819,
+ "An": 2820,
+ "urg": 2821,
+ "▁learn": 2822,
+ "▁exc": 2823,
+ "▁imp": 2824,
+ "▁means": 2825,
+ "▁wur": 2826,
+ "ends": 2827,
+ "void": 2828,
+ "▁std": 2829,
+ "▁particular": 2830,
+ "ja": 2831,
+ "▁source": 2832,
+ "default": 2833,
+ "py": 2834,
+ "▁als": 2835,
+ "scri": 2836,
+ "status": 2837,
+ "▁story": 2838,
+ "▁begin": 2839,
+ "▁position": 2840,
+ "▁special": 2841,
+ "php": 2842,
+ "▁bar": 2843,
+ "▁pract": 2844,
+ "call": 2845,
+ "▁das": 2846,
+ "▁rad": 2847,
+ "▁close": 2848,
+ "www": 2849,
+ "ере": 2850,
+ "gu": 2851,
+ "▁Er": 2852,
+ "▁dom": 2853,
+ "AM": 2854,
+ "▁bed": 2855,
+ "▁several": 2856,
+ "aul": 2857,
+ "box": 2858,
+ "▁low": 2859,
+ "pack": 2860,
+ "Reg": 2861,
+ "Of": 2862,
+ "atures": 2863,
+ "én": 2864,
+ "eder": 2865,
+ "uilder": 2866,
+ "cast": 2867,
+ "conom": 2868,
+ "raft": 2869,
+ "▁makes": 2870,
+ "Loc": 2871,
+ "http": 2872,
+ "▁abs": 2873,
+ "resh": 2874,
+ "▁Will": 2875,
+ "break": 2876,
+ "▁options": 2877,
+ "fort": 2878,
+ "▁из": 2879,
+ "▁anal": 2880,
+ "▁env": 2881,
+ "({": 2882,
+ "event": 2883,
+ "▁page": 2884,
+ "ternal": 2885,
+ "▁distribut": 2886,
+ "▁food": 2887,
+ "check": 2888,
+ "CK": 2889,
+ "▁во": 2890,
+ "assert": 2891,
+ "án": 2892,
+ "base": 2893,
+ "▁whole": 2894,
+ "ación": 2895,
+ "OD": 2896,
+ "▁turned": 2897,
+ "igma": 2898,
+ "▁response": 2899,
+ "▁University": 2900,
+ "▁div": 2901,
+ "apter": 2902,
+ "▁results": 2903,
+ "▁represent": 2904,
+ "▁everything": 2905,
+ "▁Cent": 2906,
+ "utes": 2907,
+ "rix": 2908,
+ "▁Some": 2909,
+ "▁behind": 2910,
+ "▁creat": 2911,
+ "place": 2912,
+ "su": 2913,
+ "▁Part": 2914,
+ "umb": 2915,
+ "mathbb": 2916,
+ "ping": 2917,
+ "▁match": 2918,
+ "Out": 2919,
+ "dom": 2920,
+ "▁situ": 2921,
+ "dr": 2922,
+ "ara": 2923,
+ "▁window": 2924,
+ "ns": 2925,
+ "lished": 2926,
+ "▁Ver": 2927,
+ "▁message": 2928,
+ "▁Em": 2929,
+ "▁human": 2930,
+ "perties": 2931,
+ "лу": 2932,
+ "lem": 2933,
+ "ORT": 2934,
+ "▁early": 2935,
+ "▁quick": 2936,
+ "▁та": 2937,
+ "roid": 2938,
+ "▁country": 2939,
+ "▁due": 2940,
+ "▁Die": 2941,
+ "▁trying": 2942,
+ "▁live": 2943,
+ "▁press": 2944,
+ "INT": 2945,
+ "With": 2946,
+ "oved": 2947,
+ "▁specific": 2948,
+ "▁fall": 2949,
+ "uk": 2950,
+ "yl": 2951,
+ "▁general": 2952,
+ "му": 2953,
+ "ну": 2954,
+ "▁names": 2955,
+ "where": 2956,
+ "▁These": 2957,
+ "▁sil": 2958,
+ "ét": 2959,
+ "▁ener": 2960,
+ "▁Now": 2961,
+ "▁address": 2962,
+ "Response": 2963,
+ "▁Mr": 2964,
+ "▁answ": 2965,
+ "▁film": 2966,
+ "▁strong": 2967,
+ "▁bring": 2968,
+ "▁United": 2969,
+ "▁ge": 2970,
+ "▁woman": 2971,
+ "New": 2972,
+ "ett": 2973,
+ ".)": 2974,
+ "ename": 2975,
+ "▁AN": 2976,
+ "▁describ": 2977,
+ "за": 2978,
+ "ising": 2979,
+ "EL": 2980,
+ "ql": 2981,
+ "▁fur": 2982,
+ "ying": 2983,
+ "▁Cal": 2984,
+ "▁Dr": 2985,
+ "ERR": 2986,
+ "▁\\\\": 2987,
+ "angle": 2988,
+ "urope": 2989,
+ "▁city": 2990,
+ "▁index": 2991,
+ "▁action": 2992,
+ "▁However": 2993,
+ "▁fig": 2994,
+ "ias": 2995,
+ "▁question": 2996,
+ "▁Jan": 2997,
+ "▁Med": 2998,
+ "▁Cont": 2999,
+ "amed": 3000,
+ "Call": 3001,
+ "plied": 3002,
+ "tty": 3003,
+ "▁individ": 3004,
+ "page": 3005,
+ "▁comb": 3006,
+ "section": 3007,
+ "▁Comm": 3008,
+ "uel": 3009,
+ "▁het": 3010,
+ "▁Bar": 3011,
+ "agement": 3012,
+ "fin": 3013,
+ "▁major": 3014,
+ "oper": 3015,
+ "api": 3016,
+ "room": 3017,
+ "▁„": 3018,
+ "▁hab": 3019,
+ "зи": 3020,
+ "▁auf": 3021,
+ "current": 3022,
+ "ni": 3023,
+ "▁include": 3024,
+ "▁qui": 3025,
+ "va": 3026,
+ "UE": 3027,
+ "▁idea": 3028,
+ ",'": 3029,
+ "▁required": 3030,
+ "▁heart": 3031,
+ "ibility": 3032,
+ "iction": 3033,
+ "Model": 3034,
+ "write": 3035,
+ "▁content": 3036,
+ "▁wer": 3037,
+ "▁hands": 3038,
+ "zen": 3039,
+ "char": 3040,
+ "}^{": 3041,
+ "▁mass": 3042,
+ "ply": 3043,
+ "▁nat": 3044,
+ "rel": 3045,
+ "▁dat": 3046,
+ "================": 3047,
+ "imal": 3048,
+ "▁probably": 3049,
+ "unch": 3050,
+ "▁mer": 3051,
+ "ilar": 3052,
+ "ires": 3053,
+ "▁watch": 3054,
+ "SI": 3055,
+ "▁cult": 3056,
+ "▁mother": 3057,
+ "▁government": 3058,
+ "ording": 3059,
+ "▁()": 3060,
+ "▁pri": 3061,
+ "▁link": 3062,
+ "group": 3063,
+ "OL": 3064,
+ "▁near": 3065,
+ "▁Ser": 3066,
+ "Ser": 3067,
+ "ito": 3068,
+ "▁values": 3069,
+ "▁java": 3070,
+ "fully": 3071,
+ "Count": 3072,
+ "++)": 3073,
+ "▁vi": 3074,
+ "▁white": 3075,
+ "mat": 3076,
+ "ctx": 3077,
+ "▁conc": 3078,
+ "▁stay": 3079,
+ "ging": 3080,
+ "▁clear": 3081,
+ "▁copy": 3082,
+ "selves": 3083,
+ "▁provide": 3084,
+ "▁words": 3085,
+ "comp": 3086,
+ "args": 3087,
+ "▁pick": 3088,
+ "uly": 3089,
+ "▁vari": 3090,
+ "▁believe": 3091,
+ "▁Co": 3092,
+ "Property": 3093,
+ "Group": 3094,
+ "▁ten": 3095,
+ "ischen": 3096,
+ "eturn": 3097,
+ "ival": 3098,
+ "System": 3099,
+ "CL": 3100,
+ "bed": 3101,
+ "▁total": 3102,
+ "▁ist": 3103,
+ "Input": 3104,
+ "uments": 3105,
+ "Manager": 3106,
+ "ши": 3107,
+ "▁win": 3108,
+ "leep": 3109,
+ "PI": 3110,
+ "ного": 3111,
+ "ruction": 3112,
+ "▁inte": 3113,
+ "App": 3114,
+ "avor": 3115,
+ "▁respect": 3116,
+ "ators": 3117,
+ "▁como": 3118,
+ "▁cut": 3119,
+ "FA": 3120,
+ "▁sus": 3121,
+ "▁App": 3122,
+ "rect": 3123,
+ "FI": 3124,
+ "▁began": 3125,
+ "oph": 3126,
+ "▁sort": 3127,
+ "though": 3128,
+ "је": 3129,
+ "icro": 3130,
+ "Trans": 3131,
+ "лі": 3132,
+ "▁Inst": 3133,
+ "request": 3134,
+ "ор": 3135,
+ "▁relations": 3136,
+ "-\\": 3137,
+ "Status": 3138,
+ "жи": 3139,
+ "▁father": 3140,
+ "cs": 3141,
+ "▁sex": 3142,
+ "isch": 3143,
+ "vo": 3144,
+ "}_{": 3145,
+ "aven": 3146,
+ "▁Ne": 3147,
+ "ATE": 3148,
+ "itten": 3149,
+ "▁ess": 3150,
+ "TH": 3151,
+ "ights": 3152,
+ "▁hom": 3153,
+ "▁today": 3154,
+ "▁zu": 3155,
+ "ita": 3156,
+ "▁isn": 3157,
+ "▁opt": 3158,
+ "ogn": 3159,
+ "ér": 3160,
+ "▁whether": 3161,
+ "ixed": 3162,
+ "phi": 3163,
+ "idence": 3164,
+ "ald": 3165,
+ "Client": 3166,
+ "At": 3167,
+ "▁death": 3168,
+ "▁Let": 3169,
+ "ius": 3170,
+ "ги": 3171,
+ "▁ре": 3172,
+ "ben": 3173,
+ ")\r": 3174,
+ "ba": 3175,
+ ">": 3176,
+ "avel": 3177,
+ "▁miss": 3178,
+ "▁node": 3179,
+ "▁($": 3180,
+ "▁color": 3181,
+ "▁obt": 3182,
+ "tot": 3183,
+ "▁пре": 3184,
+ "CON": 3185,
+ "ette": 3186,
+ "▁Go": 3187,
+ "Fl": 3188,
+ "▁Don": 3189,
+ "▁crit": 3190,
+ "▁ri": 3191,
+ "post": 3192,
+ "▁->": 3193,
+ "▁Just": 3194,
+ "What": 3195,
+ "atal": 3196,
+ "▁Min": 3197,
+ "▁Cor": 3198,
+ "▁dark": 3199,
+ "rl": 3200,
+ "▁larg": 3201,
+ "ding": 3202,
+ "ón": 3203,
+ "ouch": 3204,
+ "▁um": 3205,
+ "▁elect": 3206,
+ "▁dam": 3207,
+ "▁needs": 3208,
+ "▁matter": 3209,
+ "▁rather": 3210,
+ "from": 3211,
+ "ram": 3212,
+ "▁і": 3213,
+ "▁taken": 3214,
+ "▁deal": 3215,
+ "▁period": 3216,
+ "▁Mon": 3217,
+ "▁Л": 3218,
+ "▁Aug": 3219,
+ "run": 3220,
+ "mm": 3221,
+ "elle": 3222,
+ "▁export": 3223,
+ "Sc": 3224,
+ "vis": 3225,
+ "abor": 3226,
+ "▁author": 3227,
+ "ère": 3228,
+ "▁remember": 3229,
+ "▁redu": 3230,
+ "▁List": 3231,
+ "▁focus": 3232,
+ "▁character": 3233,
+ "Table": 3234,
+ "▁individual": 3235,
+ "▁needed": 3236,
+ "bum": 3237,
+ "▁style": 3238,
+ "inary": 3239,
+ "ersion": 3240,
+ "oute": 3241,
+ "▁Pe": 3242,
+ "▁hon": 3243,
+ "mut": 3244,
+ "see": 3245,
+ "▁became": 3246,
+ "▁dire": 3247,
+ "▁document": 3248,
+ "sec": 3249,
+ "ening": 3250,
+ "▁visit": 3251,
+ "▁fac": 3252,
+ "tx": 3253,
+ "down": 3254,
+ "plit": 3255,
+ "▁phys": 3256,
+ "itting": 3257,
+ "joy": 3258,
+ "▁hig": 3259,
+ "This": 3260,
+ "Ad": 3261,
+ "▁Brit": 3262,
+ "▁employ": 3263,
+ "▁ré": 3264,
+ "▁т": 3265,
+ "lambda": 3266,
+ "▁impro": 3267,
+ "▁Bo": 3268,
+ "iding": 3269,
+ "▁online": 3270,
+ "mem": 3271,
+ "atform": 3272,
+ "▁War": 3273,
+ "▁cas": 3274,
+ "asure": 3275,
+ "▁pur": 3276,
+ "medi": 3277,
+ "Dis": 3278,
+ "▁Germ": 3279,
+ "pc": 3280,
+ "са": 3281,
+ "▁friends": 3282,
+ "▁Mc": 3283,
+ "DI": 3284,
+ "▁plus": 3285,
+ "▁Set": 3286,
+ "iddle": 3287,
+ "itut": 3288,
+ "▁depend": 3289,
+ "rest": 3290,
+ "▁Je": 3291,
+ "▁hor": 3292,
+ "▁entire": 3293,
+ "Query": 3294,
+ "▁refer": 3295,
+ "▁hot": 3296,
+ "▁Aust": 3297,
+ "▁common": 3298,
+ "ці": 3299,
+ "▁pull": 3300,
+ "▁Add": 3301,
+ "▁season": 3302,
+ "▁invol": 3303,
+ "▁World": 3304,
+ "client": 3305,
+ "now": 3306,
+ "true": 3307,
+ "append": 3308,
+ "itted": 3309,
+ "empt": 3310,
+ "){": 3311,
+ "///": 3312,
+ "▁prop": 3313,
+ "imate": 3314,
+ "SC": 3315,
+ "▁hours": 3316,
+ "▁hope": 3317,
+ "andom": 3318,
+ "ід": 3319,
+ "istic": 3320,
+ "▁property": 3321,
+ "sg": 3322,
+ ">(": 3323,
+ "▁write": 3324,
+ "mark": 3325,
+ "find": 3326,
+ "▁personal": 3327,
+ "][": 3328,
+ "rown": 3329,
+ "Ph": 3330,
+ "▁foot": 3331,
+ "▁research": 3332,
+ "ironment": 3333,
+ "▁nom": 3334,
+ "▁instance": 3335,
+ "▁held": 3336,
+ "De": 3337,
+ "▁members": 3338,
+ "▁fire": 3339,
+ "▁history": 3340,
+ "▁map": 3341,
+ "▁discuss": 3342,
+ "▁espec": 3343,
+ "▁taking": 3344,
+ "▁services": 3345,
+ "▁indust": 3346,
+ "igen": 3347,
+ "▁Ass": 3348,
+ "▁expected": 3349,
+ "▁wurde": 3350,
+ "dir": 3351,
+ "▁among": 3352,
+ "▁sugg": 3353,
+ "rec": 3354,
+ "Inter": 3355,
+ "block": 3356,
+ "▁Rep": 3357,
+ "▁pain": 3358,
+ "▁five": 3359,
+ "▁fund": 3360,
+ "rid": 3361,
+ "arrow": 3362,
+ "▁treat": 3363,
+ "▁heard": 3364,
+ "▁determ": 3365,
+ "icult": 3366,
+ "▁sense": 3367,
+ "ese": 3368,
+ "Fun": 3369,
+ "▁months": 3370,
+ "json": 3371,
+ ",”": 3372,
+ "TI": 3373,
+ "orage": 3374,
+ "▁У": 3375,
+ "▁everyone": 3376,
+ "▁clos": 3377,
+ "iers": 3378,
+ "airs": 3379,
+ "define": 3380,
+ "If": 3381,
+ "osp": 3382,
+ "▁wonder": 3383,
+ "NA": 3384,
+ "query": 3385,
+ "pg": 3386,
+ "ites": 3387,
+ "▁material": 3388,
+ "yd": 3389,
+ "Read": 3390,
+ "html": 3391,
+ "TE": 3392,
+ "Pr": 3393,
+ "^{\\": 3394,
+ "▁gave": 3395,
+ "▁IS": 3396,
+ "▁suggest": 3397,
+ "Override": 3398,
+ "rodu": 3399,
+ "From": 3400,
+ "▁Europe": 3401,
+ "PO": 3402,
+ "▁soon": 3403,
+ "host": 3404,
+ "▁Ber": 3405,
+ "....": 3406,
+ "▁Har": 3407,
+ "▁energy": 3408,
+ "><": 3409,
+ "aves": 3410,
+ "▁easy": 3411,
+ "▁bre": 3412,
+ "frame": 3413,
+ "▁ground": 3414,
+ "with": 3415,
+ "▁inside": 3416,
+ "ief": 3417,
+ "▁mo": 3418,
+ "pm": 3419,
+ "pan": 3420,
+ "igr": 3421,
+ "▁om": 3422,
+ "next": 3423,
+ "omet": 3424,
+ "▁status": 3425,
+ "▁}\r": 3426,
+ "▁music": 3427,
+ "ora": 3428,
+ "iles": 3429,
+ "ki": 3430,
+ "▁esc": 3431,
+ "▁bes": 3432,
+ "▁Dis": 3433,
+ "▁host": 3434,
+ "▁comes": 3435,
+ "used": 3436,
+ "▁future": 3437,
+ "lick": 3438,
+ "aid": 3439,
+ "▁compet": 3440,
+ "▁voice": 3441,
+ "▁load": 3442,
+ "evel": 3443,
+ "▁neg": 3444,
+ "▁command": 3445,
+ "▁für": 3446,
+ "▁pie": 3447,
+ "▁quite": 3448,
+ "▁blo": 3449,
+ "agn": 3450,
+ "ilon": 3451,
+ "▁claim": 3452,
+ "▁teach": 3453,
+ "▁previous": 3454,
+ "▁site": 3455,
+ "color": 3456,
+ "attr": 3457,
+ "▁accept": 3458,
+ "▁exact": 3459,
+ ")}": 3460,
+ "aft": 3461,
+ "roller": 3462,
+ "он": 3463,
+ "oo": 3464,
+ "Date": 3465,
+ "▁ou": 3466,
+ "sy": 3467,
+ "▁pretty": 3468,
+ "▁image": 3469,
+ "BU": 3470,
+ "▁terms": 3471,
+ "▁search": 3472,
+ "▁è": 3473,
+ "▁Val": 3474,
+ "▁‘": 3475,
+ "▁Dav": 3476,
+ "MS": 3477,
+ "src": 3478,
+ "mar": 3479,
+ "incip": 3480,
+ "▁couldn": 3481,
+ "ados": 3482,
+ "▁dro": 3483,
+ "beta": 3484,
+ "imum": 3485,
+ "▁minutes": 3486,
+ "▁grand": 3487,
+ "▁»": 3488,
+ "▁Our": 3489,
+ "Str": 3490,
+ "VER": 3491,
+ "maz": 3492,
+ "▁original": 3493,
+ "ini": 3494,
+ "▁coll": 3495,
+ "loat": 3496,
+ "▁os": 3497,
+ "});": 3498,
+ "summary": 3499,
+ "▁wall": 3500,
+ "Color": 3501,
+ "▁vers": 3502,
+ "▁della": 3503,
+ "▁\"\"\"": 3504,
+ "mathbf": 3505,
+ "zer": 3506,
+ "aur": 3507,
+ "▁track": 3508,
+ "▁associ": 3509,
+ "▁suff": 3510,
+ "▁inde": 3511,
+ "ague": 3512,
+ "▁Apr": 3513,
+ "Le": 3514,
+ "roups": 3515,
+ "board": 3516,
+ "▁attack": 3517,
+ "▁series": 3518,
+ "▁instead": 3519,
+ "ham": 3520,
+ "book": 3521,
+ "▁six": 3522,
+ "▁Rec": 3523,
+ "▁coming": 3524,
+ "urt": 3525,
+ "▁global": 3526,
+ "▁necess": 3527,
+ "lege": 3528,
+ "Pos": 3529,
+ "▁leave": 3530,
+ "▁pod": 3531,
+ "ategory": 3532,
+ "uz": 3533,
+ "▁deep": 3534,
+ "▁km": 3535,
+ "▁outside": 3536,
+ "has": 3537,
+ "options": 3538,
+ "▁Sm": 3539,
+ "Sub": 3540,
+ "rows": 3541,
+ "▁ви": 3542,
+ "▁States": 3543,
+ "▁wrong": 3544,
+ "▁however": 3545,
+ "▁sem": 3546,
+ "▁catch": 3547,
+ "\"),": 3548,
+ "model": 3549,
+ "▁http": 3550,
+ "▁option": 3551,
+ "rie": 3552,
+ "▁ста": 3553,
+ "▁är": 3554,
+ "▁enjoy": 3555,
+ "nu": 3556,
+ "▁pas": 3557,
+ "▁amount": 3558,
+ "▁respons": 3559,
+ "▁Intern": 3560,
+ "▁myself": 3561,
+ "▁opp": 3562,
+ "▁Sim": 3563,
+ "▁sens": 3564,
+ "Ed": 3565,
+ "▁(\\": 3566,
+ "▁students": 3567,
+ "нов": 3568,
+ "▁points": 3569,
+ "arning": 3570,
+ "UP": 3571,
+ "elling": 3572,
+ "▁cannot": 3573,
+ "Be": 3574,
+ "▁length": 3575,
+ "null": 3576,
+ "uint": 3577,
+ "wise": 3578,
+ "▁double": 3579,
+ "ige": 3580,
+ "ista": 3581,
+ "▁estab": 3582,
+ "anch": 3583,
+ "▁ago": 3584,
+ "▁bound": 3585,
+ "▁fa": 3586,
+ "▁clean": 3587,
+ "▁simple": 3588,
+ "mi": 3589,
+ "########": 3590,
+ "ifier": 3591,
+ "▁General": 3592,
+ "▁seemed": 3593,
+ "ena": 3594,
+ "▁age": 3595,
+ "ной": 3596,
+ "endif": 3597,
+ "AA": 3598,
+ "▁caus": 3599,
+ "▁educ": 3600,
+ "▁cell": 3601,
+ "Gener": 3602,
+ "space": 3603,
+ "▁Your": 3604,
+ "▁beaut": 3605,
+ "gt": 3606,
+ "▁limit": 3607,
+ "▁date": 3608,
+ "Util": 3609,
+ "▁National": 3610,
+ "ows": 3611,
+ "pat": 3612,
+ "quad": 3613,
+ "▁ok": 3614,
+ "▁И": 3615,
+ "arth": 3616,
+ "hat": 3617,
+ "▁community": 3618,
+ "oul": 3619,
+ "▁econom": 3620,
+ "Component": 3621,
+ "bor": 3622,
+ "usion": 3623,
+ "▁below": 3624,
+ "earch": 3625,
+ "ores": 3626,
+ "ban": 3627,
+ "▁August": 3628,
+ "▁further": 3629,
+ "sigma": 3630,
+ "▁ha": 3631,
+ "ji": 3632,
+ "▁comput": 3633,
+ "гра": 3634,
+ "▁None": 3635,
+ "▁ter": 3636,
+ "▁anyone": 3637,
+ "▁task": 3638,
+ "ente": 3639,
+ "position": 3640,
+ "pped": 3641,
+ "▁aus": 3642,
+ "Attribute": 3643,
+ "req": 3644,
+ "addr": 3645,
+ "light": 3646,
+ "ше": 3647,
+ "▁arm": 3648,
+ "cover": 3649,
+ "upport": 3650,
+ "▁Gl": 3651,
+ "▁San": 3652,
+ "▁writing": 3653,
+ "▁lost": 3654,
+ "▁Mark": 3655,
+ "▁gre": 3656,
+ "TYPE": 3657,
+ "▁South": 3658,
+ "▁perfect": 3659,
+ "▁package": 3660,
+ "▁infl": 3661,
+ "haps": 3662,
+ "▁Ang": 3663,
+ "respon": 3664,
+ "ris": 3665,
+ "ptember": 3666,
+ "▁building": 3667,
+ "VAL": 3668,
+ "free": 3669,
+ "▁ce": 3670,
+ "HT": 3671,
+ "▁From": 3672,
+ "ds": 3673,
+ "roy": 3674,
+ "achine": 3675,
+ "nown": 3676,
+ "▁saying": 3677,
+ "▁бы": 3678,
+ "oe": 3679,
+ "Ref": 3680,
+ "▁network": 3681,
+ "parent": 3682,
+ "uge": 3683,
+ "▁similar": 3684,
+ ">\r": 3685,
+ "Builder": 3686,
+ "▁living": 3687,
+ "▁continue": 3688,
+ "anger": 3689,
+ "▁Red": 3690,
+ "▁hair": 3691,
+ "anced": 3692,
+ "ians": 3693,
+ "▁dead": 3694,
+ "▁boolean": 3695,
+ "ication": 3696,
+ "▁де": 3697,
+ "▁client": 3698,
+ "uct": 3699,
+ "▁•": 3700,
+ "SP": 3701,
+ "older": 3702,
+ "пе": 3703,
+ "udio": 3704,
+ "▁deg": 3705,
+ "asing": 3706,
+ "▁step": 3707,
+ "▁pers": 3708,
+ "ção": 3709,
+ "obj": 3710,
+ "oz": 3711,
+ "ula": 3712,
+ "▁round": 3713,
+ "▁upon": 3714,
+ "▁resource": 3715,
+ "▁valid": 3716,
+ "▁II": 3717,
+ "bug": 3718,
+ "std": 3719,
+ "▁ang": 3720,
+ "span": 3721,
+ "pol": 3722,
+ "ialog": 3723,
+ "▁phot": 3724,
+ "?'": 3725,
+ "DB": 3726,
+ "▁Fin": 3727,
+ "VE": 3728,
+ "Em": 3729,
+ "▁cam": 3730,
+ "target": 3731,
+ "pected": 3732,
+ "Hel": 3733,
+ "▁ut": 3734,
+ "▁Test": 3735,
+ "▁town": 3736,
+ "align": 3737,
+ "▁webs": 3738,
+ "inner": 3739,
+ "augh": 3740,
+ "▁except": 3741,
+ "▁initial": 3742,
+ "enty": 3743,
+ "lich": 3744,
+ "▁Aut": 3745,
+ "top": 3746,
+ "▁fail": 3747,
+ "ona": 3748,
+ "▁benef": 3749,
+ "anks": 3750,
+ "ische": 3751,
+ ".*": 3752,
+ "▁signific": 3753,
+ "▁contact": 3754,
+ "Rec": 3755,
+ "ario": 3756,
+ "ottom": 3757,
+ "▁relationship": 3758,
+ "]);": 3759,
+ "▁На": 3760,
+ "Head": 3761,
+ "format": 3762,
+ "▁ét": 3763,
+ "▁More": 3764,
+ "actory": 3765,
+ "portun": 3766,
+ "+\\": 3767,
+ "▁simply": 3768,
+ "▁ep": 3769,
+ "▁Russ": 3770,
+ "ní": 3771,
+ "ua": 3772,
+ "erc": 3773,
+ "▁longer": 3774,
+ "inition": 3775,
+ "ector": 3776,
+ "aption": 3777,
+ "▁profess": 3778,
+ "▁Mus": 3779,
+ "ilities": 3780,
+ "ès": 3781,
+ "▁Act": 3782,
+ "offset": 3783,
+ "▁ill": 3784,
+ "band": 3785,
+ "▁Ag": 3786,
+ "▁По": 3787,
+ "би": 3788,
+ "content": 3789,
+ "icon": 3790,
+ "▁works": 3791,
+ "ynam": 3792,
+ "plement": 3793,
+ "Resource": 3794,
+ "Action": 3795,
+ "▁difficult": 3796,
+ "▁West": 3797,
+ "▁video": 3798,
+ "▁THE": 3799,
+ "▁decl": 3800,
+ "ondon": 3801,
+ "ded": 3802,
+ "}{\\": 3803,
+ "ocr": 3804,
+ "▁City": 3805,
+ "▁я": 3806,
+ "uer": 3807,
+ "cz": 3808,
+ "▁imag": 3809,
+ "cr": 3810,
+ "ete": 3811,
+ "idget": 3812,
+ "▁Mod": 3813,
+ "▁forward": 3814,
+ "▁pict": 3815,
+ "orge": 3816,
+ "▁subject": 3817,
+ "update": 3818,
+ "attle": 3819,
+ "sa": 3820,
+ "▁Ant": 3821,
+ "▁running": 3822,
+ "▁sal": 3823,
+ "conne": 3824,
+ "▁output": 3825,
+ "adata": 3826,
+ "ML": 3827,
+ "Check": 3828,
+ "ledge": 3829,
+ "▁paper": 3830,
+ "params": 3831,
+ "avy": 3832,
+ "▁af": 3833,
+ "▁eine": 3834,
+ "▁jour": 3835,
+ "AY": 3836,
+ "▁itself": 3837,
+ "▁Str": 3838,
+ "style": 3839,
+ "That": 3840,
+ "▁million": 3841,
+ "▁language": 3842,
+ "OS": 3843,
+ "ving": 3844,
+ "▁ма": 3845,
+ "▁то": 3846,
+ ")(": 3847,
+ "▁buy": 3848,
+ "./": 3849,
+ "▁...": 3850,
+ "▁tried": 3851,
+ "▁compl": 3852,
+ "▁activ": 3853,
+ "apped": 3854,
+ "Button": 3855,
+ "Token": 3856,
+ "▁provided": 3857,
+ "iber": 3858,
+ "▁created": 3859,
+ "curity": 3860,
+ "End": 3861,
+ "ał": 3862,
+ "uster": 3863,
+ "izing": 3864,
+ "omb": 3865,
+ "▁sich": 3866,
+ "▁compon": 3867,
+ "▁See": 3868,
+ "▁uint": 3869,
+ "▁label": 3870,
+ "vol": 3871,
+ "ów": 3872,
+ "ocol": 3873,
+ "▁received": 3874,
+ "▁intern": 3875,
+ "це": 3876,
+ "Run": 3877,
+ "▁road": 3878,
+ "▁Oct": 3879,
+ "▁Comp": 3880,
+ "▁study": 3881,
+ "▁те": 3882,
+ "Act": 3883,
+ "▁tour": 3884,
+ "▁State": 3885,
+ "▁added": 3886,
+ "https": 3887,
+ "stream": 3888,
+ "▁lower": 3889,
+ "▁box": 3890,
+ "▁Sk": 3891,
+ "▁themselves": 3892,
+ "▁cross": 3893,
+ "▁echo": 3894,
+ "▁device": 3895,
+ "pose": 3896,
+ "▁games": 3897,
+ "PL": 3898,
+ "Window": 3899,
+ "ises": 3900,
+ "title": 3901,
+ "Stream": 3902,
+ "zt": 3903,
+ "▁Sw": 3904,
+ "▁role": 3905,
+ "iant": 3906,
+ "ku": 3907,
+ "sequ": 3908,
+ "▁late": 3909,
+ "▁sold": 3910,
+ "ря": 3911,
+ "Comm": 3912,
+ "▁entre": 3913,
+ "▁dog": 3914,
+ "device": 3915,
+ "Par": 3916,
+ "▁likely": 3917,
+ "^{-": 3918,
+ "▁len": 3919,
+ "▁Paul": 3920,
+ "▁tool": 3921,
+ "Off": 3922,
+ "▁famil": 3923,
+ "▁draw": 3924,
+ "apping": 3925,
+ "▁events": 3926,
+ "cret": 3927,
+ "rought": 3928,
+ "Content": 3929,
+ "▁software": 3930,
+ "ria": 3931,
+ "msg": 3932,
+ "gamma": 3933,
+ "▁hear": 3934,
+ "Oper": 3935,
+ "▁yourself": 3936,
+ "▁liter": 3937,
+ "emp": 3938,
+ "▁separ": 3939,
+ "▁З": 3940,
+ "▁title": 3941,
+ "Method": 3942,
+ "mathrm": 3943,
+ "▁slow": 3944,
+ "▁Rom": 3945,
+ "!!": 3946,
+ "▁tax": 3947,
+ "ска": 3948,
+ "emplate": 3949,
+ "oi": 3950,
+ "▁Art": 3951,
+ "false": 3952,
+ "astic": 3953,
+ "сть": 3954,
+ "ocket": 3955,
+ "▁ens": 3956,
+ "TO": 3957,
+ "amente": 3958,
+ "local": 3959,
+ "chie": 3960,
+ "▁pan": 3961,
+ "ний": 3962,
+ "chema": 3963,
+ "▁North": 3964,
+ "зо": 3965,
+ "▁>=": 3966,
+ "Aut": 3967,
+ "▁dig": 3968,
+ "▁seems": 3969,
+ "▁morning": 3970,
+ "sole": 3971,
+ "umer": 3972,
+ "delta": 3973,
+ "ité": 3974,
+ "abase": 3975,
+ "raf": 3976,
+ "▁observ": 3977,
+ "▁Est": 3978,
+ "▁seg": 3979,
+ "▁[]": 3980,
+ "▁Pres": 3981,
+ "iful": 3982,
+ "push": 3983,
+ "▁Off": 3984,
+ "ipe": 3985,
+ "ati": 3986,
+ "▁dim": 3987,
+ "ceed": 3988,
+ "Ent": 3989,
+ "____": 3990,
+ "entry": 3991,
+ "▁fight": 3992,
+ "▁cred": 3993,
+ "▁OR": 3994,
+ "▁Dep": 3995,
+ "${": 3996,
+ "лен": 3997,
+ "Create": 3998,
+ "▁April": 3999,
+ "ministr": 4000,
+ "FL": 4001,
+ "▁Ap": 4002,
+ "▁Here": 4003,
+ "private": 4004,
+ "Instance": 4005,
+ "iem": 4006,
+ "▁office": 4007,
+ "▁third": 4008,
+ "▁update": 4009,
+ "Line": 4010,
+ "tag": 4011,
+ "▁especially": 4012,
+ "▁года": 4013,
+ "▁cu": 4014,
+ "▁kill": 4015,
+ "aught": 4016,
+ "▁swe": 4017,
+ "Options": 4018,
+ "IM": 4019,
+ "CC": 4020,
+ "▁compan": 4021,
+ "just": 4022,
+ "▁While": 4023,
+ "izer": 4024,
+ "▁мо": 4025,
+ "ке": 4026,
+ "▁auto": 4027,
+ "▁band": 4028,
+ "мен": 4029,
+ "iques": 4030,
+ "▁ple": 4031,
+ "NO": 4032,
+ "▁OF": 4033,
+ "▁song": 4034,
+ "▁Acc": 4035,
+ "EXT": 4036,
+ "ensor": 4037,
+ "ining": 4038,
+ "▁lat": 4039,
+ "big": 4040,
+ "▁King": 4041,
+ "och": 4042,
+ "si": 4043,
+ "▁Hist": 4044,
+ "▁quality": 4045,
+ "mode": 4046,
+ "▁opportun": 4047,
+ "▁wouldn": 4048,
+ ":**": 4049,
+ "output": 4050,
+ "▁feet": 4051,
+ "▁mis": 4052,
+ "df": 4053,
+ "aging": 4054,
+ "▁ме": 4055,
+ "▁tro": 4056,
+ "▁defined": 4057,
+ "▁review": 4058,
+ "▁Fil": 4059,
+ ">>": 4060,
+ "▁princip": 4061,
+ "Base": 4062,
+ "dict": 4063,
+ "verage": 4064,
+ "icient": 4065,
+ "IF": 4066,
+ "▁hit": 4067,
+ "Page": 4068,
+ "▁perm": 4069,
+ "cel": 4070,
+ "ít": 4071,
+ "▁express": 4072,
+ "▁indic": 4073,
+ "▁September": 4074,
+ "image": 4075,
+ "▁products": 4076,
+ "▁media": 4077,
+ "change": 4078,
+ "igger": 4079,
+ "▁send": 4080,
+ "last": 4081,
+ "ming": 4082,
+ "pa": 4083,
+ "uary": 4084,
+ "▁speak": 4085,
+ "ный": 4086,
+ "ще": 4087,
+ "ysis": 4088,
+ "lying": 4089,
+ "▁ч": 4090,
+ "like": 4091,
+ "ры": 4092,
+ "ві": 4093,
+ "▁Mich": 4094,
+ "MO": 4095,
+ "▁Jah": 4096,
+ "ensive": 4097,
+ "▁share": 4098,
+ "▁development": 4099,
+ "CP": 4100,
+ "spec": 4101,
+ "▁fast": 4102,
+ "het": 4103,
+ "HO": 4104,
+ "▁particip": 4105,
+ "Block": 4106,
+ "▁viol": 4107,
+ "▁frame": 4108,
+ "▁qual": 4109,
+ "tre": 4110,
+ "▁Ф": 4111,
+ "▁toward": 4112,
+ "fg": 4113,
+ "Box": 4114,
+ "Column": 4115,
+ "▁milit": 4116,
+ "▁March": 4117,
+ "▁various": 4118,
+ "pass": 4119,
+ "▁Park": 4120,
+ "▁Ben": 4121,
+ "Frame": 4122,
+ "▁normal": 4123,
+ "open": 4124,
+ "px": 4125,
+ "▁phone": 4126,
+ "▁Even": 4127,
+ "▁ma": 4128,
+ "ibrary": 4129,
+ "Start": 4130,
+ "idden": 4131,
+ "rho": 4132,
+ "graph": 4133,
+ "acing": 4134,
+ "'.": 4135,
+ "arter": 4136,
+ "mes": 4137,
+ "inst": 4138,
+ "▁ir": 4139,
+ "active": 4140,
+ "▁fem": 4141,
+ "▁moved": 4142,
+ "▁store": 4143,
+ "▁price": 4144,
+ "\").": 4145,
+ "berg": 4146,
+ "▁nov": 4147,
+ "▁card": 4148,
+ "ellow": 4149,
+ "▁party": 4150,
+ "▁Mor": 4151,
+ "ael": 4152,
+ "▁percent": 4153,
+ "▁training": 4154,
+ "▁ing": 4155,
+ "imer": 4156,
+ "▁Sam": 4157,
+ "Default": 4158,
+ "▁fuck": 4159,
+ "▁complete": 4160,
+ "uid": 4161,
+ "▁details": 4162,
+ "▁led": 4163,
+ "Point": 4164,
+ "▁Count": 4165,
+ "▁regard": 4166,
+ "zo": 4167,
+ "▁Bro": 4168,
+ "▁recogn": 4169,
+ "▁Hol": 4170,
+ "UM": 4171,
+ "element": 4172,
+ "Mode": 4173,
+ "▁exam": 4174,
+ "▁EX": 4175,
+ "Image": 4176,
+ "verse": 4177,
+ "riter": 4178,
+ "soft": 4179,
+ "▁introdu": 4180,
+ "▁surpr": 4181,
+ "Buffer": 4182,
+ "lector": 4183,
+ "aren": 4184,
+ "anged": 4185,
+ "▁Pat": 4186,
+ "▁Pal": 4187,
+ "▁contr": 4188,
+ "Handler": 4189,
+ "▁features": 4190,
+ "iple": 4191,
+ "▁CON": 4192,
+ "Fil": 4193,
+ "▁Port": 4194,
+ "▁thinking": 4195,
+ "doc": 4196,
+ "wer": 4197,
+ "▁worked": 4198,
+ "PC": 4199,
+ "cm": 4200,
+ "dat": 4201,
+ "PRO": 4202,
+ "▁Every": 4203,
+ "▁era": 4204,
+ "▁First": 4205,
+ "gn": 4206,
+ "▁immedi": 4207,
+ "ovember": 4208,
+ "apan": 4209,
+ "▁extra": 4210,
+ "▁section": 4211,
+ "▁June": 4212,
+ "▁via": 4213,
+ "▁gone": 4214,
+ "come": 4215,
+ "▁stri": 4216,
+ "^\\": 4217,
+ "antly": 4218,
+ "▁arch": 4219,
+ "Source": 4220,
+ "▁conv": 4221,
+ "▁London": 4222,
+ "Number": 4223,
+ "▁questions": 4224,
+ "andid": 4225,
+ "▁played": 4226,
+ "env": 4227,
+ "▁School": 4228,
+ "▁natural": 4229,
+ "can": 4230,
+ "▁news": 4231,
+ "DR": 4232,
+ "▁chall": 4233,
+ "▁Soc": 4234,
+ "▁э": 4235,
+ "▁attempt": 4236,
+ "*}": 4237,
+ "Null": 4238,
+ "rote": 4239,
+ "▁bi": 4240,
+ "▁written": 4241,
+ "▁blood": 4242,
+ "▁happened": 4243,
+ "▁cause": 4244,
+ "ashing": 4245,
+ "▁William": 4246,
+ "adem": 4247,
+ "▁brought": 4248,
+ "▁display": 4249,
+ "ima": 4250,
+ "▁finally": 4251,
+ "tab": 4252,
+ "▁returned": 4253,
+ "ных": 4254,
+ "nie": 4255,
+ "▁q": 4256,
+ "▁hers": 4257,
+ "▁Pre": 4258,
+ "▁dou": 4259,
+ "buffer": 4260,
+ "▁effort": 4261,
+ "aine": 4262,
+ "xy": 4263,
+ "▁histor": 4264,
+ "enu": 4265,
+ "▁arriv": 4266,
+ "▁Dem": 4267,
+ "▁favor": 4268,
+ "▁handle": 4269,
+ "SET": 4270,
+ "▁Public": 4271,
+ "rupt": 4272,
+ "▁ur": 4273,
+ "▁force": 4274,
+ "▁és": 4275,
+ "ube": 4276,
+ "Pre": 4277,
+ "рі": 4278,
+ "iny": 4279,
+ "theta": 4280,
+ "isf": 4281,
+ "▁national": 4282,
+ "Equal": 4283,
+ "rench": 4284,
+ "▁wife": 4285,
+ "▁capt": 4286,
+ "▁Inter": 4287,
+ "tau": 4288,
+ "▁sleep": 4289,
+ "../../": 4290,
+ "▁issue": 4291,
+ "▁member": 4292,
+ "▁await": 4293,
+ "▁Dan": 4294,
+ "zi": 4295,
+ "inate": 4296,
+ "▁sym": 4297,
+ "chan": 4298,
+ "▁Jack": 4299,
+ "▁English": 4300,
+ "▁sz": 4301,
+ "ributes": 4302,
+ "▁ign": 4303,
+ "ál": 4304,
+ "▁appear": 4305,
+ "rad": 4306,
+ "idge": 4307,
+ "▁couple": 4308,
+ "▁ship": 4309,
+ "lig": 4310,
+ "web": 4311,
+ "▁usually": 4312,
+ "▁ready": 4313,
+ "▁vill": 4314,
+ "▁Why": 4315,
+ "ebru": 4316,
+ "▁grad": 4317,
+ "ords": 4318,
+ "▁inf": 4319,
+ "▁loss": 4320,
+ "▁od": 4321,
+ "▁Phil": 4322,
+ "server": 4323,
+ "▁Up": 4324,
+ "▁buff": 4325,
+ "▁filename": 4326,
+ "ABLE": 4327,
+ "iting": 4328,
+ "efore": 4329,
+ "()->": 4330,
+ "▁conditions": 4331,
+ "vm": 4332,
+ "eld": 4333,
+ "itz": 4334,
+ "▁Trans": 4335,
+ "▁weight": 4336,
+ "▁higher": 4337,
+ "▁rate": 4338,
+ "▁accom": 4339,
+ "vider": 4340,
+ "OM": 4341,
+ "▁ways": 4342,
+ "coming": 4343,
+ "▁lock": 4344,
+ "▁etc": 4345,
+ "▁avec": 4346,
+ "▁takes": 4347,
+ "▁Char": 4348,
+ "▁November": 4349,
+ "method": 4350,
+ "▁Austral": 4351,
+ "▁America": 4352,
+ "long": 4353,
+ "cember": 4354,
+ "▁political": 4355,
+ "flow": 4356,
+ "▁maybe": 4357,
+ "▁amb": 4358,
+ "Layout": 4359,
+ "iled": 4360,
+ "omen": 4361,
+ "ola": 4362,
+ "icip": 4363,
+ "partial": 4364,
+ "True": 4365,
+ "▁floor": 4366,
+ "▁Def": 4367,
+ "▁concern": 4368,
+ "yr": 4369,
+ "▁shows": 4370,
+ "ih": 4371,
+ "▁answer": 4372,
+ "acc": 4373,
+ "▁ball": 4374,
+ "▁Rev": 4375,
+ "▁sun": 4376,
+ "▁quickly": 4377,
+ "▁somet": 4378,
+ "mente": 4379,
+ "▁Mal": 4380,
+ "undred": 4381,
+ "▁issues": 4382,
+ "ecause": 4383,
+ "pes": 4384,
+ "▁player": 4385,
+ "▁parents": 4386,
+ "▁popular": 4387,
+ "▁mode": 4388,
+ "▁mention": 4389,
+ "NE": 4390,
+ "Load": 4391,
+ "▁regular": 4392,
+ "aved": 4393,
+ "?:": 4394,
+ "year": 4395,
+ "func": 4396,
+ "▁performance": 4397,
+ "▁July": 4398,
+ "thern": 4399,
+ "▁website": 4400,
+ "ford": 4401,
+ "PR": 4402,
+ "ela": 4403,
+ "level": 4404,
+ "uit": 4405,
+ "flags": 4406,
+ "▁worth": 4407,
+ "▁correspon": 4408,
+ "▁British": 4409,
+ "sim": 4410,
+ "▁alone": 4411,
+ "▁har": 4412,
+ "▁ones": 4413,
+ "obile": 4414,
+ "▁dru": 4415,
+ "chi": 4416,
+ "▁David": 4417,
+ "▁problems": 4418,
+ "▁column": 4419,
+ "();\r": 4420,
+ "ZE": 4421,
+ "▁relig": 4422,
+ "ological": 4423,
+ "▁region": 4424,
+ "ady": 4425,
+ "IO": 4426,
+ "ander": 4427,
+ "Net": 4428,
+ "▁built": 4429,
+ "▁install": 4430,
+ "▁approach": 4431,
+ "Cur": 4432,
+ "▁fine": 4433,
+ "▁talking": 4434,
+ "▁changes": 4435,
+ "Style": 4436,
+ "▁Mart": 4437,
+ "лю": 4438,
+ "response": 4439,
+ "teger": 4440,
+ "{\r": 4441,
+ "irit": 4442,
+ "▁protected": 4443,
+ "▁rele": 4444,
+ "ership": 4445,
+ "тель": 4446,
+ "unsigned": 4447,
+ "ialize": 4448,
+ "▁https": 4449,
+ "Tag": 4450,
+ "▁$(": 4451,
+ "more": 4452,
+ "ypes": 4453,
+ "▁stream": 4454,
+ "etch": 4455,
+ "▁engine": 4456,
+ "KE": 4457,
+ "cmd": 4458,
+ "script": 4459,
+ "ttp": 4460,
+ "▁avoid": 4461,
+ "▁terr": 4462,
+ "▁rock": 4463,
+ "▁ful": 4464,
+ "Update": 4465,
+ "▁environment": 4466,
+ "▁prec": 4467,
+ "▁са": 4468,
+ "▁cases": 4469,
+ "▁offset": 4470,
+ "▁rais": 4471,
+ "lib": 4472,
+ "ées": 4473,
+ "aa": 4474,
+ "yt": 4475,
+ "▁arr": 4476,
+ "opyright": 4477,
+ "first": 4478,
+ "▁util": 4479,
+ "▁feature": 4480,
+ "posed": 4481,
+ "ffect": 4482,
+ "жа": 4483,
+ "itude": 4484,
+ "ements": 4485,
+ "asc": 4486,
+ "ador": 4487,
+ "lections": 4488,
+ "▁club": 4489,
+ "]{": 4490,
+ "▁*)": 4491,
+ "ство": 4492,
+ "▁imm": 4493,
+ "▁former": 4494,
+ "▁rights": 4495,
+ "▁decided": 4496,
+ "▁rev": 4497,
+ "▁ment": 4498,
+ "ani": 4499,
+ "▁stru": 4500,
+ "▁attention": 4501,
+ "artment": 4502,
+ "▁Ital": 4503,
+ "alle": 4504,
+ "▁bis": 4505,
+ "gener": 4506,
+ "▁integr": 4507,
+ "ello": 4508,
+ "rypt": 4509,
+ "▁achie": 4510,
+ "nes": 4511,
+ "▁stra": 4512,
+ "sb": 4513,
+ "▁types": 4514,
+ "▁RE": 4515,
+ "Init": 4516,
+ "▁comment": 4517,
+ "▁addition": 4518,
+ "▁ID": 4519,
+ "ART": 4520,
+ "FO": 4521,
+ "щи": 4522,
+ "Conne": 4523,
+ "▁squ": 4524,
+ "▁considered": 4525,
+ "idad": 4526,
+ "▁October": 4527,
+ "cial": 4528,
+ "▁Of": 4529,
+ "▁travel": 4530,
+ "▁boy": 4531,
+ "').": 4532,
+ "uy": 4533,
+ "illa": 4534,
+ "istry": 4535,
+ "▁va": 4536,
+ "▁Che": 4537,
+ "ERT": 4538,
+ "ende": 4539,
+ "ungen": 4540,
+ "aby": 4541,
+ "▁Rober": 4542,
+ "▁playing": 4543,
+ "ils": 4544,
+ "▁sam": 4545,
+ "▁execut": 4546,
+ "▁Us": 4547,
+ "▁mut": 4548,
+ "▁bal": 4549,
+ "asse": 4550,
+ "▁kids": 4551,
+ "▁financ": 4552,
+ "gor": 4553,
+ "▁Sec": 4554,
+ "bert": 4555,
+ "▁High": 4556,
+ "▁је": 4557,
+ "▁kept": 4558,
+ "button": 4559,
+ "itory": 4560,
+ "▁Rem": 4561,
+ "▁DE": 4562,
+ "▁reach": 4563,
+ "▁bur": 4564,
+ "Label": 4565,
+ "át": 4566,
+ "ago": 4567,
+ "▁passed": 4568,
+ "▁behav": 4569,
+ "xFF": 4570,
+ "▁Return": 4571,
+ "STR": 4572,
+ "▁Les": 4573,
+ "▁ord": 4574,
+ "ala": 4575,
+ "inger": 4576,
+ "▁Since": 4577,
+ "▁experi": 4578,
+ "▁shall": 4579,
+ "▁star": 4580,
+ "non": 4581,
+ "▁gun": 4582,
+ "▁Bel": 4583,
+ "▁obj": 4584,
+ "ares": 4585,
+ "rs": 4586,
+ "▁weeks": 4587,
+ "nen": 4588,
+ "▁Stre": 4589,
+ "oring": 4590,
+ "▁î": 4591,
+ "▁serious": 4592,
+ "times": 4593,
+ "▁House": 4594,
+ "▁roll": 4595,
+ "▁register": 4596,
+ "▁module": 4597,
+ "▁applic": 4598,
+ "IR": 4599,
+ "▁cook": 4600,
+ "aux": 4601,
+ "▁save": 4602,
+ "▁Cr": 4603,
+ ",\r": 4604,
+ "▁states": 4605,
+ "▁empty": 4606,
+ "▁autom": 4607,
+ "figure": 4608,
+ "iance": 4609,
+ "▁happy": 4610,
+ "▁fn": 4611,
+ "▁jud": 4612,
+ "▁hat": 4613,
+ "ACK": 4614,
+ "▁Fe": 4615,
+ "$-": 4616,
+ "ivil": 4617,
+ "oted": 4618,
+ "▁sizeof": 4619,
+ "▁situation": 4620,
+ "▁lives": 4621,
+ "▁feeling": 4622,
+ "▁risk": 4623,
+ "▁January": 4624,
+ "▁Object": 4625,
+ "▁recomm": 4626,
+ "▁вы": 4627,
+ "▁potential": 4628,
+ "eah": 4629,
+ "▁complex": 4630,
+ "printf": 4631,
+ "istance": 4632,
+ "irth": 4633,
+ "lik": 4634,
+ "aste": 4635,
+ "▁whose": 4636,
+ "Arg": 4637,
+ "▁modern": 4638,
+ "iones": 4639,
+ "▁че": 4640,
+ "▁sett": 4641,
+ "▁Mag": 4642,
+ "ae": 4643,
+ "▁condition": 4644,
+ "Length": 4645,
+ "▁fit": 4646,
+ "ounds": 4647,
+ "▁changed": 4648,
+ "▁guy": 4649,
+ "filter": 4650,
+ "atever": 4651,
+ "éd": 4652,
+ "remove": 4653,
+ "▁hop": 4654,
+ "▁Out": 4655,
+ "▁Rich": 4656,
+ "child": 4657,
+ "▁included": 4658,
+ "$\\": 4659,
+ "▁Tom": 4660,
+ "eline": 4661,
+ "▁sometimes": 4662,
+ "▁drink": 4663,
+ "▁quant": 4664,
+ "▁please": 4665,
+ "▁Int": 4666,
+ "rief": 4667,
+ "▁exactly": 4668,
+ "cing": 4669,
+ "▁allowed": 4670,
+ "build": 4671,
+ "▁beautiful": 4672,
+ "▁Well": 4673,
+ "▁looks": 4674,
+ "▁ü": 4675,
+ "▁chance": 4676,
+ "▁wrote": 4677,
+ "▁nor": 4678,
+ "▁failed": 4679,
+ "Met": 4680,
+ "▁prior": 4681,
+ "▁hundred": 4682,
+ "ской": 4683,
+ "oria": 4684,
+ "▁cy": 4685,
+ "▁web": 4686,
+ "▁mess": 4687,
+ "leq": 4688,
+ "dy": 4689,
+ "tex": 4690,
+ "▁anim": 4691,
+ "atur": 4692,
+ "▁structure": 4693,
+ "option": 4694,
+ "▁actual": 4695,
+ "▁Franc": 4696,
+ "enced": 4697,
+ ".": 4698,
+ "▁flow": 4699,
+ "▁Afr": 4700,
+ "det": 4701,
+ "▁Ke": 4702,
+ "ety": 4703,
+ "ский": 4704,
+ "▁stuff": 4705,
+ "itter": 4706,
+ "▁args": 4707,
+ "▁album": 4708,
+ "▁]": 4709,
+ "ugin": 4710,
+ "SU": 4711,
+ "Per": 4712,
+ "▁circ": 4713,
+ "▁correct": 4714,
+ "▁lines": 4715,
+ "▁completely": 4716,
+ "known": 4717,
+ "▁tree": 4718,
+ "root": 4719,
+ "▁Japan": 4720,
+ "oles": 4721,
+ "endo": 4722,
+ "▁location": 4723,
+ "▁Х": 4724,
+ "▁mid": 4725,
+ "aling": 4726,
+ "GL": 4727,
+ "iano": 4728,
+ "▁{}": 4729,
+ "lang": 4730,
+ "▁equip": 4731,
+ "ERROR": 4732,
+ "▁memory": 4733,
+ "▁(\"": 4734,
+ "▁nature": 4735,
+ "google": 4736,
+ "abs": 4737,
+ "BC": 4738,
+ "▁gets": 4739,
+ "Command": 4740,
+ "TER": 4741,
+ "aled": 4742,
+ "cp": 4743,
+ "▁purch": 4744,
+ "▁Den": 4745,
+ "▁herself": 4746,
+ "▁Ir": 4747,
+ "▁sie": 4748,
+ "gar": 4749,
+ "Ap": 4750,
+ "▁nel": 4751,
+ "ota": 4752,
+ ")]": 4753,
+ "cor": 4754,
+ "acht": 4755,
+ "(*": 4756,
+ "irtual": 4757,
+ "▁police": 4758,
+ "▁skin": 4759,
+ "ship": 4760,
+ "efined": 4761,
+ "aughter": 4762,
+ "inding": 4763,
+ "▁Sl": 4764,
+ "▁influ": 4765,
+ "▁mount": 4766,
+ "▁az": 4767,
+ "▁wood": 4768,
+ "otes": 4769,
+ "ega": 4770,
+ "▁according": 4771,
+ "▁namespace": 4772,
+ "Delta": 4773,
+ "stant": 4774,
+ "▁published": 4775,
+ "aker": 4776,
+ "▁Black": 4777,
+ "ln": 4778,
+ "▁industry": 4779,
+ "SON": 4780,
+ "Rep": 4781,
+ "▁choice": 4782,
+ "▁inn": 4783,
+ "kl": 4784,
+ "▁pal": 4785,
+ "▁aud": 4786,
+ "▁standard": 4787,
+ "▁knowledge": 4788,
+ "**,": 4789,
+ "▁Frank": 4790,
+ "sq": 4791,
+ "Output": 4792,
+ "▁för": 4793,
+ "Valid": 4794,
+ "ugh": 4795,
+ "▁books": 4796,
+ "▁James": 4797,
+ "ko": 4798,
+ "▁companies": 4799,
+ "anning": 4800,
+ "▁vict": 4801,
+ "▁repl": 4802,
+ "▁sche": 4803,
+ "▁happen": 4804,
+ "fty": 4805,
+ "acity": 4806,
+ "ira": 4807,
+ "▁implement": 4808,
+ "ского": 4809,
+ "number": 4810,
+ "SH": 4811,
+ "iro": 4812,
+ "▁fear": 4813,
+ "▁touch": 4814,
+ "▁cast": 4815,
+ "ASS": 4816,
+ "▁consist": 4817,
+ "Task": 4818,
+ "▁sig": 4819,
+ "ба": 4820,
+ "igation": 4821,
+ "▁Most": 4822,
+ "▁Der": 4823,
+ "}(\\": 4824,
+ ":\"": 4825,
+ "▁Fig": 4826,
+ "ali": 4827,
+ "iner": 4828,
+ "'),": 4829,
+ "▁Coun": 4830,
+ "(_": 4831,
+ "▁distributed": 4832,
+ "NAME": 4833,
+ "▁mur": 4834,
+ "▁career": 4835,
+ "~~": 4836,
+ "pers": 4837,
+ "aries": 4838,
+ "enses": 4839,
+ "▁Also": 4840,
+ "Version": 4841,
+ "▁unique": 4842,
+ "▁France": 4843,
+ "BA": 4844,
+ "ky": 4845,
+ "▁Febru": 4846,
+ "▁died": 4847,
+ "omega": 4848,
+ "▁Form": 4849,
+ "▁width": 4850,
+ "tocol": 4851,
+ "▁lie": 4852,
+ "She": 4853,
+ "ém": 4854,
+ "▁straight": 4855,
+ "▁nach": 4856,
+ "▁stood": 4857,
+ "olds": 4858,
+ "▁goes": 4859,
+ "cell": 4860,
+ "▁till": 4861,
+ "LI": 4862,
+ "draw": 4863,
+ "▁satisf": 4864,
+ "▁reading": 4865,
+ "ATION": 4866,
+ "▁Are": 4867,
+ "▁Ac": 4868,
+ ")*": 4869,
+ "▁additional": 4870,
+ "wood": 4871,
+ "cil": 4872,
+ "пу": 4873,
+ "ULT": 4874,
+ "▁bill": 4875,
+ "mas": 4876,
+ "ania": 4877,
+ "су": 4878,
+ "anz": 4879,
+ "height": 4880,
+ "jo": 4881,
+ "▁dos": 4882,
+ "\\\"": 4883,
+ "▁/>": 4884,
+ "▁production": 4885,
+ "iger": 4886,
+ "▁ст": 4887,
+ "show": 4888,
+ "▁population": 4889,
+ "▁park": 4890,
+ "▁Ze": 4891,
+ "▁necessary": 4892,
+ "▁trust": 4893,
+ "▁shown": 4894,
+ "module": 4895,
+ "GE": 4896,
+ "▁lay": 4897,
+ "▁announ": 4898,
+ "▁className": 4899,
+ "▁calcul": 4900,
+ "Function": 4901,
+ "▁Sal": 4902,
+ "OK": 4903,
+ "TP": 4904,
+ "▁entry": 4905,
+ "▁Stud": 4906,
+ "▁items": 4907,
+ "▁security": 4908,
+ "Entry": 4909,
+ "float": 4910,
+ "ls": 4911,
+ "ibly": 4912,
+ "▁contribut": 4913,
+ "▁Check": 4914,
+ "MD": 4915,
+ "▁improve": 4916,
+ "Part": 4917,
+ "▁systems": 4918,
+ "Bl": 4919,
+ "▁policy": 4920,
+ "▁screen": 4921,
+ "▁Any": 4922,
+ "▁opened": 4923,
+ "alloc": 4924,
+ "▁December": 4925,
+ "▁É": 4926,
+ "▁email": 4927,
+ "ader": 4928,
+ "=>": 4929,
+ "▁Hen": 4930,
+ "▁info": 4931,
+ "▁float": 4932,
+ "▁switch": 4933,
+ "ран": 4934,
+ "urance": 4935,
+ "▁assum": 4936,
+ "ustr": 4937,
+ "▁groups": 4938,
+ "▁Read": 4939,
+ "▁wat": 4940,
+ "Sp": 4941,
+ "вер": 4942,
+ "RAN": 4943,
+ "hib": 4944,
+ "ALL": 4945,
+ "▁hus": 4946,
+ "Spec": 4947,
+ "\"))": 4948,
+ "▁French": 4949,
+ "▁Class": 4950,
+ "▁president": 4951,
+ "▁definit": 4952,
+ "▁Nor": 4953,
+ "▁Thom": 4954,
+ "aign": 4955,
+ "Width": 4956,
+ "Do": 4957,
+ "▁{@": 4958,
+ "agon": 4959,
+ "▁Lu": 4960,
+ "▁followed": 4961,
+ "MM": 4962,
+ "asons": 4963,
+ "tmp": 4964,
+ "▁throws": 4965,
+ "ITY": 4966,
+ "ном": 4967,
+ "▁fair": 4968,
+ "▁pen": 4969,
+ "ég": 4970,
+ "▁interface": 4971,
+ "▁saf": 4972,
+ "oon": 4973,
+ "Back": 4974,
+ "▁speed": 4975,
+ "▁extends": 4976,
+ "empty": 4977,
+ "▁пере": 4978,
+ "▁proper": 4979,
+ "▁driv": 4980,
+ "фи": 4981,
+ "▁center": 4982,
+ "header": 4983,
+ "▁})": 4984,
+ "wa": 4985,
+ "▁middle": 4986,
+ "▁choose": 4987,
+ "▁Stad": 4988,
+ "SO": 4989,
+ "Factory": 4990,
+ "Dev": 4991,
+ "icles": 4992,
+ "▁application": 4993,
+ "▁models": 4994,
+ "pite": 4995,
+ "cap": 4996,
+ "xi": 4997,
+ "ospital": 4998,
+ "▁dream": 4999,
+ "END": 5000,
+ "▁contract": 5001,
+ "icrosoft": 5002,
+ "▁thous": 5003,
+ "izes": 5004,
+ "▁да": 5005,
+ "▁CO": 5006,
+ "▁direction": 5007,
+ "▁``": 5008,
+ "▁drive": 5009,
+ "Max": 5010,
+ "cia": 5011,
+ "▁continu": 5012,
+ "▁Alex": 5013,
+ "▁gold": 5014,
+ "▁prep": 5015,
+ "▁origin": 5016,
+ "▁rap": 5017,
+ "Op": 5018,
+ "ously": 5019,
+ "▁areas": 5020,
+ "PORT": 5021,
+ "она": 5022,
+ "▁safe": 5023,
+ "▁professional": 5024,
+ "apache": 5025,
+ "▁temper": 5026,
+ "sz": 5027,
+ "▁unit": 5028,
+ "▁cop": 5029,
+ "eqn": 5030,
+ "Listener": 5031,
+ "▁format": 5032,
+ "select": 5033,
+ "▁comfort": 5034,
+ "▁meant": 5035,
+ "iday": 5036,
+ "eme": 5037,
+ "▁active": 5038,
+ "▁note": 5039,
+ "▁Mil": 5040,
+ "only": 5041,
+ "▁<=": 5042,
+ "▁neigh": 5043,
+ "ao": 5044,
+ "▁blue": 5045,
+ "▁TV": 5046,
+ "Child": 5047,
+ "▁reached": 5048,
+ "Address": 5049,
+ "ств": 5050,
+ "▁closed": 5051,
+ "inder": 5052,
+ "olo": 5053,
+ "▁alt": 5054,
+ "▁adm": 5055,
+ "Format": 5056,
+ "UI": 5057,
+ "▁Ham": 5058,
+ "▁frequ": 5059,
+ "▁independ": 5060,
+ "▁easily": 5061,
+ "▁Land": 5062,
+ "▁tor": 5063,
+ "ography": 5064,
+ "infty": 5065,
+ "▁Work": 5066,
+ "iven": 5067,
+ "▁County": 5068,
+ "▁src": 5069,
+ "}$,": 5070,
+ "parse": 5071,
+ "CD": 5072,
+ "▁Cour": 5073,
+ "▁fol": 5074,
+ "Entity": 5075,
+ "pgf": 5076,
+ "▁China": 5077,
+ "▁Sub": 5078,
+ "hood": 5079,
+ "▁fields": 5080,
+ "▁yes": 5081,
+ "rend": 5082,
+ "▁towards": 5083,
+ "▁staff": 5084,
+ "▁Air": 5085,
+ "▁station": 5086,
+ "atives": 5087,
+ "▁impact": 5088,
+ "вы": 5089,
+ "▁directly": 5090,
+ "issions": 5091,
+ "iva": 5092,
+ "|\\": 5093,
+ "Ptr": 5094,
+ "▁Sant": 5095,
+ "Pol": 5096,
+ "▁progress": 5097,
+ "itar": 5098,
+ "▁parts": 5099,
+ "▁plant": 5100,
+ "▁absolut": 5101,
+ "▁guess": 5102,
+ "eqref": 5103,
+ "▁tim": 5104,
+ "▁Lou": 5105,
+ "▁cool": 5106,
+ "alu": 5107,
+ "▁mouth": 5108,
+ "них": 5109,
+ "▁height": 5110,
+ "gest": 5111,
+ "▁Post": 5112,
+ "▁board": 5113,
+ "▁tit": 5114,
+ "▁hour": 5115,
+ "▁server": 5116,
+ "▁players": 5117,
+ "rier": 5118,
+ "Link": 5119,
+ "▁President": 5120,
+ "](": 5121,
+ "▁construct": 5122,
+ "handle": 5123,
+ "}$.": 5124,
+ "rying": 5125,
+ "▁shop": 5126,
+ "iana": 5127,
+ "exp": 5128,
+ "Helper": 5129,
+ "Offset": 5130,
+ "aches": 5131,
+ "▁connection": 5132,
+ "▁difference": 5133,
+ "service": 5134,
+ "▁gas": 5135,
+ "▁priv": 5136,
+ "▁univers": 5137,
+ "▁wish": 5138,
+ "Rem": 5139,
+ "Url": 5140,
+ "geb": 5141,
+ "So": 5142,
+ "ensions": 5143,
+ "Module": 5144,
+ "SIZE": 5145,
+ "▁prem": 5146,
+ "window": 5147,
+ "▁dies": 5148,
+ "del": 5149,
+ "▁row": 5150,
+ "▁average": 5151,
+ "xim": 5152,
+ "▁pu": 5153,
+ "anç": 5154,
+ "Det": 5155,
+ "ker": 5156,
+ "ya": 5157,
+ "▁Det": 5158,
+ "▁på": 5159,
+ "▁named": 5160,
+ "▁decision": 5161,
+ "win": 5162,
+ "▁George": 5163,
+ "arily": 5164,
+ "▁solution": 5165,
+ "▁multiple": 5166,
+ "ategy": 5167,
+ "▁learning": 5168,
+ "▁secret": 5169,
+ "DO": 5170,
+ "▁nice": 5171,
+ "////////////////": 5172,
+ "Su": 5173,
+ "itation": 5174,
+ "▁join": 5175,
+ "▁elements": 5176,
+ "▁emer": 5177,
+ "tilde": 5178,
+ "▁dep": 5179,
+ "▁shot": 5180,
+ "▁platform": 5181,
+ "othing": 5182,
+ "My": 5183,
+ "edia": 5184,
+ "oms": 5185,
+ "aily": 5186,
+ "([": 5187,
+ "▁dress": 5188,
+ "▁official": 5189,
+ "estern": 5190,
+ "▁discover": 5191,
+ "▁mi": 5192,
+ "ные": 5193,
+ "CA": 5194,
+ "oding": 5195,
+ "▁Found": 5196,
+ "▁affect": 5197,
+ "Vis": 5198,
+ "stract": 5199,
+ "iced": 5200,
+ "debug": 5201,
+ "▁related": 5202,
+ "▁spect": 5203,
+ "ushed": 5204,
+ "сько": 5205,
+ "▁bank": 5206,
+ "▁cele": 5207,
+ "AND": 5208,
+ "olf": 5209,
+ "ем": 5210,
+ "▁fill": 5211,
+ "▁gives": 5212,
+ "▁бу": 5213,
+ "aron": 5214,
+ "▁Jes": 5215,
+ "REG": 5216,
+ "▁sudd": 5217,
+ "dated": 5218,
+ "vi": 5219,
+ "▁gi": 5220,
+ "send": 5221,
+ "cpp": 5222,
+ "▁spent": 5223,
+ "ande": 5224,
+ "▁operation": 5225,
+ "process": 5226,
+ "▁inform": 5227,
+ "▁Free": 5228,
+ "yond": 5229,
+ "▁perhaps": 5230,
+ "▁surv": 5231,
+ "▁Loc": 5232,
+ "▁concl": 5233,
+ "▁раз": 5234,
+ "▁Over": 5235,
+ "hol": 5236,
+ "raz": 5237,
+ "Write": 5238,
+ "▁giving": 5239,
+ "rd": 5240,
+ "instance": 5241,
+ "▁released": 5242,
+ "▁Ro": 5243,
+ "RA": 5244,
+ "▁practice": 5245,
+ "▁graph": 5246,
+ "▁increase": 5247,
+ "▁figure": 5248,
+ "Filter": 5249,
+ "HECK": 5250,
+ "idx": 5251,
+ "▁glass": 5252,
+ "ski": 5253,
+ "comes": 5254,
+ "▁cat": 5255,
+ "▁cold": 5256,
+ "goto": 5257,
+ "ufact": 5258,
+ "▁Copyright": 5259,
+ "}}\\": 5260,
+ "▁streng": 5261,
+ "▁dir": 5262,
+ "token": 5263,
+ "▁occur": 5264,
+ "arlier": 5265,
+ "▁measure": 5266,
+ "▁sec": 5267,
+ "▁más": 5268,
+ "▁Net": 5269,
+ "▁argument": 5270,
+ "▁sou": 5271,
+ "▁moving": 5272,
+ "▁prefer": 5273,
+ "mask": 5274,
+ "<<": 5275,
+ "▁breath": 5276,
+ "▁physical": 5277,
+ "▁positive": 5278,
+ "▁sor": 5279,
+ "▁depart": 5280,
+ "▁remove": 5281,
+ "▁kit": 5282,
+ "▁meeting": 5283,
+ "▁Data": 5284,
+ "ograf": 5285,
+ "actions": 5286,
+ "▁parameters": 5287,
+ "▁Att": 5288,
+ "esch": 5289,
+ "▁involved": 5290,
+ "ät": 5291,
+ "LL": 5292,
+ "Bar": 5293,
+ "▁си": 5294,
+ "ech": 5295,
+ "GET": 5296,
+ "▁prevent": 5297,
+ "▁beyond": 5298,
+ "▁Other": 5299,
+ "än": 5300,
+ "byte": 5301,
+ "▁sudden": 5302,
+ "olve": 5303,
+ "▁но": 5304,
+ "LOG": 5305,
+ "unit": 5306,
+ "▁truth": 5307,
+ "rat": 5308,
+ "SD": 5309,
+ "▁eat": 5310,
+ "▁Mad": 5311,
+ "▁provides": 5312,
+ "▁session": 5313,
+ "Dele": 5314,
+ "▁convers": 5315,
+ "center": 5316,
+ "▁continued": 5317,
+ "otion": 5318,
+ "cache": 5319,
+ "display": 5320,
+ "▁protect": 5321,
+ "ams": 5322,
+ "▁pow": 5323,
+ "CTION": 5324,
+ "▁Mac": 5325,
+ "mo": 5326,
+ "ха": 5327,
+ "▁distance": 5328,
+ "▁Time": 5329,
+ "gi": 5330,
+ "▁sequ": 5331,
+ "Target": 5332,
+ "сле": 5333,
+ "Server": 5334,
+ "▁wide": 5335,
+ "close": 5336,
+ "▁cru": 5337,
+ "Ext": 5338,
+ "▁select": 5339,
+ "▁pattern": 5340,
+ "\"));": 5341,
+ "Provider": 5342,
+ "URL": 5343,
+ "▁green": 5344,
+ "▁waiting": 5345,
+ "proto": 5346,
+ "▁immediately": 5347,
+ "common": 5348,
+ "azione": 5349,
+ "river": 5350,
+ "▁sen": 5351,
+ "▁!==": 5352,
+ "▁February": 5353,
+ "urb": 5354,
+ "▁Sen": 5355,
+ "dest": 5356,
+ "": 5357,
+ "▁edge": 5358,
+ "▁mais": 5359,
+ "gorith": 5360,
+ "cpu": 5361,
+ "▁education": 5362,
+ "▁associated": 5363,
+ "None": 5364,
+ "hi": 5365,
+ "▁poor": 5366,
+ "sem": 5367,
+ "▁Wil": 5368,
+ "▁bud": 5369,
+ "▁auch": 5370,
+ "eller": 5371,
+ "▁Life": 5372,
+ "▁files": 5373,
+ "▁leading": 5374,
+ "▁obtain": 5375,
+ "▁Jul": 5376,
+ "atory": 5377,
+ "гу": 5378,
+ "itable": 5379,
+ "▁onto": 5380,
+ "▁born": 5381,
+ "orem": 5382,
+ "▁Street": 5383,
+ "▁maint": 5384,
+ "Params": 5385,
+ "rip": 5386,
+ "▁ST": 5387,
+ "uv": 5388,
+ "main": 5389,
+ "▁▁▁▁▁▁▁": 5390,
+ "▁recent": 5391,
+ "Web": 5392,
+ "ova": 5393,
+ "ца": 5394,
+ "aise": 5395,
+ "yles": 5396,
+ "▁described": 5397,
+ "▁beginning": 5398,
+ "▁Day": 5399,
+ "▁Vol": 5400,
+ "▁huge": 5401,
+ "Has": 5402,
+ "ancy": 5403,
+ "Header": 5404,
+ "▁aren": 5405,
+ "ван": 5406,
+ "▁ensure": 5407,
+ "▁pet": 5408,
+ "mult": 5409,
+ "▁Like": 5410,
+ "▁management": 5411,
+ "PS": 5412,
+ "while": 5413,
+ "▁background": 5414,
+ "ounter": 5415,
+ "bool": 5416,
+ "FC": 5417,
+ "Num": 5418,
+ "RL": 5419,
+ "▁excl": 5420,
+ "▁eye": 5421,
+ "img": 5422,
+ "▁rom": 5423,
+ "▁Hel": 5424,
+ "Option": 5425,
+ "▁stopped": 5426,
+ "▁thread": 5427,
+ "totype": 5428,
+ ")))": 5429,
+ "▁stage": 5430,
+ "▁über": 5431,
+ "▁although": 5432,
+ "Types": 5433,
+ "▁Oh": 5434,
+ "▁eight": 5435,
+ "▁description": 5436,
+ "''": 5437,
+ "ön": 5438,
+ "▁surface": 5439,
+ "▁International": 5440,
+ "▁charg": 5441,
+ "▁collection": 5442,
+ "▁users": 5443,
+ "▁obvious": 5444,
+ "▁century": 5445,
+ "icks": 5446,
+ "▁article": 5447,
+ "▁\"\\": 5448,
+ "dim": 5449,
+ "▁sin": 5450,
+ "enge": 5451,
+ "Control": 5452,
+ "▁commit": 5453,
+ "ensity": 5454,
+ "▁tra": 5455,
+ "criptor": 5456,
+ "▁NOT": 5457,
+ "well": 5458,
+ "▁Michael": 5459,
+ "▁nod": 5460,
+ "▁mort": 5461,
+ "ivo": 5462,
+ "isation": 5463,
+ "▁Po": 5464,
+ "▁Paris": 5465,
+ "▁administr": 5466,
+ "burg": 5467,
+ "cdot": 5468,
+ "▁military": 5469,
+ "▁Best": 5470,
+ "▁Ка": 5471,
+ "INE": 5472,
+ "▁throughout": 5473,
+ "Sl": 5474,
+ "▁impl": 5475,
+ "control": 5476,
+ "▁Ч": 5477,
+ "▁uit": 5478,
+ "▁unsigned": 5479,
+ "▁Mary": 5480,
+ "Char": 5481,
+ "мі": 5482,
+ "▁threat": 5483,
+ "▁court": 5484,
+ "ville": 5485,
+ "▁ш": 5486,
+ "▁Cam": 5487,
+ ".\r": 5488,
+ "▁currently": 5489,
+ "rot": 5490,
+ "▁Date": 5491,
+ "▁shit": 5492,
+ "▁${\\": 5493,
+ "unn": 5494,
+ "Us": 5495,
+ "▁buffer": 5496,
+ "▁sont": 5497,
+ "▁letter": 5498,
+ "inated": 5499,
+ "Change": 5500,
+ "▁href": 5501,
+ "▁lack": 5502,
+ "▁oil": 5503,
+ "▁Cons": 5504,
+ "▁Jer": 5505,
+ "BUG": 5506,
+ "iforn": 5507,
+ "▁properties": 5508,
+ "▁random": 5509,
+ "▁brother": 5510,
+ "▁piece": 5511,
+ "бу": 5512,
+ "istics": 5513,
+ "▁technology": 5514,
+ "global": 5515,
+ "▁transform": 5516,
+ "erd": 5517,
+ "▁Because": 5518,
+ "PECT": 5519,
+ "pret": 5520,
+ "▁году": 5521,
+ "▁Met": 5522,
+ "▁psy": 5523,
+ "▁од": 5524,
+ "▁god": 5525,
+ "▁Del": 5526,
+ "based": 5527,
+ "▁voor": 5528,
+ "▁Call": 5529,
+ "SA": 5530,
+ "▁filter": 5531,
+ "▁includes": 5532,
+ "olutions": 5533,
+ "fd": 5534,
+ "▁wind": 5535,
+ "▁бо": 5536,
+ "▁ability": 5537,
+ "card": 5538,
+ "▁numer": 5539,
+ "address": 5540,
+ "▁goal": 5541,
+ "ashington": 5542,
+ "▁slight": 5543,
+ "aba": 5544,
+ "▁Log": 5545,
+ "Settings": 5546,
+ "adow": 5547,
+ "▁pi": 5548,
+ "iring": 5549,
+ "FT": 5550,
+ "▁numbers": 5551,
+ "conf": 5552,
+ "task": 5553,
+ "▁în": 5554,
+ "ты": 5555,
+ "▁receive": 5556,
+ "▁root": 5557,
+ "▁India": 5558,
+ "patch": 5559,
+ "él": 5560,
+ "▁summer": 5561,
+ "▁methods": 5562,
+ "▁places": 5563,
+ "▁Ма": 5564,
+ "▁capital": 5565,
+ "▁evidence": 5566,
+ "▁German": 5567,
+ "\\,": 5568,
+ "DA": 5569,
+ "ecute": 5570,
+ "column": 5571,
+ "▁functions": 5572,
+ "▁counter": 5573,
+ "▁arms": 5574,
+ "▁feed": 5575,
+ "vey": 5576,
+ "hent": 5577,
+ "MAX": 5578,
+ "▁acqu": 5579,
+ "▁apply": 5580,
+ "▁husband": 5581,
+ "▁killed": 5582,
+ "▁Spec": 5583,
+ "entity": 5584,
+ "▁earlier": 5585,
+ "▁Miss": 5586,
+ "▁setting": 5587,
+ "itect": 5588,
+ "▁ded": 5589,
+ "Row": 5590,
+ "▁ran": 5591,
+ "▁Yes": 5592,
+ "▁financial": 5593,
+ "session": 5594,
+ "lear": 5595,
+ "ishing": 5596,
+ "▁nearly": 5597,
+ "▁dur": 5598,
+ "▁machine": 5599,
+ "xff": 5600,
+ "bro": 5601,
+ "▁symbol": 5602,
+ "lands": 5603,
+ "Acc": 5604,
+ "di": 5605,
+ "▁Robert": 5606,
+ "prop": 5607,
+ "urity": 5608,
+ "▁#####": 5609,
+ "▁walked": 5610,
+ "▁international": 5611,
+ "▁Е": 5612,
+ "Yes": 5613,
+ "▁release": 5614,
+ "▁starting": 5615,
+ "static": 5616,
+ "▁bei": 5617,
+ "allow": 5618,
+ "▁People": 5619,
+ "ez": 5620,
+ "▁parameter": 5621,
+ "Cache": 5622,
+ "▁$$": 5623,
+ "ampions": 5624,
+ "▁Mer": 5625,
+ "▁kom": 5626,
+ "leted": 5627,
+ "ois": 5628,
+ "▁Open": 5629,
+ "types": 5630,
+ "▁fue": 5631,
+ "acters": 5632,
+ "▁reference": 5633,
+ "Equals": 5634,
+ "▁aware": 5635,
+ "▁hol": 5636,
+ "▁demand": 5637,
+ "lor": 5638,
+ "▁veh": 5639,
+ "▁notice": 5640,
+ "▁component": 5641,
+ "fn": 5642,
+ "▁analysis": 5643,
+ "match": 5644,
+ "▁effective": 5645,
+ "product": 5646,
+ "ник": 5647,
+ "▁legal": 5648,
+ "ей": 5649,
+ "semb": 5650,
+ "▁located": 5651,
+ "▁су": 5652,
+ "QL": 5653,
+ "inct": 5654,
+ "eto": 5655,
+ "Draw": 5656,
+ "▁scale": 5657,
+ "ров": 5658,
+ "▁wants": 5659,
+ "How": 5660,
+ "▁wel": 5661,
+ "isions": 5662,
+ "▁deliver": 5663,
+ "under": 5664,
+ "▁deb": 5665,
+ "▁ju": 5666,
+ "values": 5667,
+ "▁sister": 5668,
+ "ков": 5669,
+ "▁Create": 5670,
+ "▁Inc": 5671,
+ "▁aux": 5672,
+ "▁White": 5673,
+ "Menu": 5674,
+ "aud": 5675,
+ "resource": 5676,
+ "▁cab": 5677,
+ "▁lif": 5678,
+ "▁culture": 5679,
+ "iche": 5680,
+ "▁whatever": 5681,
+ "▁designed": 5682,
+ "▁repe": 5683,
+ "▁Mont": 5684,
+ "▁charge": 5685,
+ "Names": 5686,
+ "▁insp": 5687,
+ "▁customers": 5688,
+ "osa": 5689,
+ "▁daughter": 5690,
+ "▁East": 5691,
+ "EQ": 5692,
+ "▁opin": 5693,
+ "▁Fre": 5694,
+ "▁seek": 5695,
+ "▁push": 5696,
+ "▁nav": 5697,
+ "▁burn": 5698,
+ "arden": 5699,
+ "hash": 5700,
+ "▁opportunity": 5701,
+ "▁Mat": 5702,
+ "oyal": 5703,
+ "▁pun": 5704,
+ "scale": 5705,
+ "ynamic": 5706,
+ "▁Type": 5707,
+ "iling": 5708,
+ "▁query": 5709,
+ "▁mist": 5710,
+ "ror": 5711,
+ "force": 5712,
+ "▁Once": 5713,
+ "▁medical": 5714,
+ "lie": 5715,
+ "▁student": 5716,
+ "ederal": 5717,
+ "▁lov": 5718,
+ "iform": 5719,
+ "▁altern": 5720,
+ "bin": 5721,
+ "oder": 5722,
+ "▁returns": 5723,
+ "register": 5724,
+ "uts": 5725,
+ "CI": 5726,
+ "▁Tor": 5727,
+ "CR": 5728,
+ "▁Los": 5729,
+ "amily": 5730,
+ "aire": 5731,
+ "++;": 5732,
+ "Controller": 5733,
+ "wide": 5734,
+ "xx": 5735,
+ "rowser": 5736,
+ "▁Book": 5737,
+ "Container": 5738,
+ "pload": 5739,
+ "▁Ev": 5740,
+ "▁tal": 5741,
+ "▁theory": 5742,
+ "eqnarray": 5743,
+ "бе": 5744,
+ "▁reported": 5745,
+ "▁meaning": 5746,
+ "▁sy": 5747,
+ "ribe": 5748,
+ "icate": 5749,
+ "hold": 5750,
+ "▁offers": 5751,
+ "▁templ": 5752,
+ "css": 5753,
+ "▁picture": 5754,
+ "▁async": 5755,
+ "▁stock": 5756,
+ "▁internal": 5757,
+ "ti": 5758,
+ "BO": 5759,
+ "Ver": 5760,
+ "спо": 5761,
+ "▁demon": 5762,
+ "▁laugh": 5763,
+ "▁End": 5764,
+ "▁kon": 5765,
+ "▁ideas": 5766,
+ "▁candid": 5767,
+ "Mem": 5768,
+ "izz": 5769,
+ "refix": 5770,
+ "▁AND": 5771,
+ "egen": 5772,
+ "El": 5773,
+ "▁campaign": 5774,
+ "Http": 5775,
+ "▁Rob": 5776,
+ "ді": 5777,
+ "▁bul": 5778,
+ "▁Ко": 5779,
+ "▁countries": 5780,
+ "».": 5781,
+ "▁expression": 5782,
+ "▁England": 5783,
+ "sf": 5784,
+ "▁certainly": 5785,
+ "agen": 5786,
+ "▁ча": 5787,
+ "▁ANY": 5788,
+ "▁connect": 5789,
+ "FE": 5790,
+ "▁android": 5791,
+ "▁Gold": 5792,
+ "▁oppos": 5793,
+ "overn": 5794,
+ "▁Commun": 5795,
+ ",_": 5796,
+ "asion": 5797,
+ "La": 5798,
+ "▁firm": 5799,
+ "▁Although": 5800,
+ "▁Good": 5801,
+ "▁Law": 5802,
+ "erve": 5803,
+ "▁brand": 5804,
+ "Min": 5805,
+ "fill": 5806,
+ "'],": 5807,
+ "▁Jew": 5808,
+ "iler": 5809,
+ "ingle": 5810,
+ "ithub": 5811,
+ "▁Div": 5812,
+ "▁cert": 5813,
+ "Height": 5814,
+ "rael": 5815,
+ "There": 5816,
+ "itute": 5817,
+ "▁amaz": 5818,
+ "look": 5819,
+ "▁SE": 5820,
+ "▁jo": 5821,
+ "▁pulled": 5822,
+ "▁resources": 5823,
+ "▁Max": 5824,
+ "▁agreed": 5825,
+ "asy": 5826,
+ "▁treatment": 5827,
+ "\">": 5828,
+ "ман": 5829,
+ "▁Err": 5830,
+ "orig": 5831,
+ "cos": 5832,
+ "▁Maybe": 5833,
+ "otal": 5834,
+ "▁train": 5835,
+ "▁Service": 5836,
+ "▁ih": 5837,
+ "▁spirit": 5838,
+ "Comp": 5839,
+ "sqrt": 5840,
+ "▁broad": 5841,
+ "}[": 5842,
+ "▁shape": 5843,
+ "▁doc": 5844,
+ "how": 5845,
+ "▁tag": 5846,
+ "atalog": 5847,
+ "sd": 5848,
+ "▁meas": 5849,
+ "▁Ро": 5850,
+ "▁exception": 5851,
+ "▁Tw": 5852,
+ "▁interesting": 5853,
+ "ATA": 5854,
+ "▁Rel": 5855,
+ "ár": 5856,
+ "▁useful": 5857,
+ "useum": 5858,
+ "▁bottom": 5859,
+ "▁otherwise": 5860,
+ "▁agree": 5861,
+ "cht": 5862,
+ "then": 5863,
+ "▁significant": 5864,
+ "}/": 5865,
+ "▁channel": 5866,
+ "icial": 5867,
+ "тив": 5868,
+ "vare": 5869,
+ "▁enter": 5870,
+ "Eng": 5871,
+ "uj": 5872,
+ "URE": 5873,
+ "queue": 5874,
+ "ono": 5875,
+ "▁contains": 5876,
+ "MI": 5877,
+ "▁nation": 5878,
+ "▁rules": 5879,
+ "fol": 5880,
+ "▁pa": 5881,
+ "arp": 5882,
+ "▁quiet": 5883,
+ "▁thus": 5884,
+ "ipped": 5885,
+ "annot": 5886,
+ "udes": 5887,
+ "():": 5888,
+ "names": 5889,
+ "▁compos": 5890,
+ "▁inj": 5891,
+ "una": 5892,
+ "bind": 5893,
+ "▁fully": 5894,
+ "ras": 5895,
+ "Utils": 5896,
+ "anges": 5897,
+ "dule": 5898,
+ "▁Christian": 5899,
+ "▁reve": 5900,
+ "änd": 5901,
+ "▁collect": 5902,
+ "▁celebr": 5903,
+ "anda": 5904,
+ "ín": 5905,
+ "join": 5906,
+ "▁paid": 5907,
+ "Core": 5908,
+ "Ge": 5909,
+ ".$": 5910,
+ "▁fif": 5911,
+ "▁uma": 5912,
+ "▁~": 5913,
+ "ervices": 5914,
+ "▁recently": 5915,
+ "desc": 5916,
+ "▁heavy": 5917,
+ "▁rule": 5918,
+ "▁Please": 5919,
+ "psi": 5920,
+ "▁console": 5921,
+ "▁fort": 5922,
+ ".\\": 5923,
+ "▁Washington": 5924,
+ "▁gar": 5925,
+ "▁Group": 5926,
+ "▁interview": 5927,
+ "anned": 5928,
+ "sql": 5929,
+ "▁anc": 5930,
+ "ја": 5931,
+ "Pack": 5932,
+ "▁Club": 5933,
+ "▁mask": 5934,
+ "▁concept": 5935,
+ "▁['": 5936,
+ "▁selected": 5937,
+ "▁Use": 5938,
+ "▁ele": 5939,
+ "ears": 5940,
+ "▁race": 5941,
+ "hy": 5942,
+ "Om": 5943,
+ "▁steps": 5944,
+ "ila": 5945,
+ "ests": 5946,
+ "eds": 5947,
+ "▁street": 5948,
+ "ners": 5949,
+ "▁birth": 5950,
+ "pop": 5951,
+ "▁ли": 5952,
+ "MB": 5953,
+ "кра": 5954,
+ "cir": 5955,
+ "epsilon": 5956,
+ "▁constant": 5957,
+ "ques": 5958,
+ "adas": 5959,
+ "▁knows": 5960,
+ "▁Py": 5961,
+ "cles": 5962,
+ "▁cit": 5963,
+ "▁pair": 5964,
+ "inese": 5965,
+ "▁Peter": 5966,
+ "▁finished": 5967,
+ "▁master": 5968,
+ "▁twenty": 5969,
+ "▁fell": 5970,
+ "▁central": 5971,
+ "▁mes": 5972,
+ "rev": 5973,
+ "STAT": 5974,
+ "stat": 5975,
+ "▁allows": 5976,
+ "▁gro": 5977,
+ "Click": 5978,
+ "▁stories": 5979,
+ "Fe": 5980,
+ "år": 5981,
+ "▁baby": 5982,
+ "encia": 5983,
+ "▁einer": 5984,
+ "Are": 5985,
+ "ebug": 5986,
+ "store": 5987,
+ "\",\"": 5988,
+ "lam": 5989,
+ "▁sv": 5990,
+ "ции": 5991,
+ "NULL": 5992,
+ "▁Leg": 5993,
+ "▁movie": 5994,
+ "▁hous": 5995,
+ "▁learned": 5996,
+ "bon": 5997,
+ "▁transfer": 5998,
+ "ifornia": 5999,
+ "psilon": 6000,
+ "▁Soft": 6001,
+ "▁commer": 6002,
+ "▁hadn": 6003,
+ "▁Ein": 6004,
+ "▁Two": 6005,
+ "craft": 6006,
+ "Process": 6007,
+ "▁под": 6008,
+ "argin": 6009,
+ "▁estim": 6010,
+ "▁Mem": 6011,
+ "ika": 6012,
+ "▁Tod": 6013,
+ "duc": 6014,
+ "▁danger": 6015,
+ "rive": 6016,
+ "Don": 6017,
+ "▁Que": 6018,
+ "hal": 6019,
+ "▁mm": 6020,
+ "▁Sur": 6021,
+ "Order": 6022,
+ "▁distribution": 6023,
+ "fa": 6024,
+ "▁Many": 6025,
+ "plicit": 6026,
+ "Empty": 6027,
+ "Handle": 6028,
+ "▁token": 6029,
+ "▁epis": 6030,
+ "▁assist": 6031,
+ "▁purpose": 6032,
+ "▁ц": 6033,
+ "NU": 6034,
+ "iders": 6035,
+ "rate": 6036,
+ "They": 6037,
+ "Parameter": 6038,
+ "Dec": 6039,
+ "▁strugg": 6040,
+ "▁shoot": 6041,
+ "IV": 6042,
+ "▁Great": 6043,
+ "▁Sil": 6044,
+ "▁loved": 6045,
+ "▁click": 6046,
+ "▁reserv": 6047,
+ "▁ве": 6048,
+ "▁spread": 6049,
+ "▁og": 6050,
+ "▁${": 6051,
+ "▁miles": 6052,
+ "▁successful": 6053,
+ "oj": 6054,
+ "▁Direct": 6055,
+ "▁ax": 6056,
+ "▁growth": 6057,
+ "Work": 6058,
+ "▁church": 6059,
+ "Inst": 6060,
+ "ICE": 6061,
+ "sten": 6062,
+ "род": 6063,
+ "▁Center": 6064,
+ "ses": 6065,
+ "got": 6066,
+ "delete": 6067,
+ "▁Ma": 6068,
+ "%%": 6069,
+ "▁crow": 6070,
+ "DF": 6071,
+ "front": 6072,
+ "▁blog": 6073,
+ "▁computer": 6074,
+ "ная": 6075,
+ "▁mir": 6076,
+ "▁Super": 6077,
+ "','": 6078,
+ "▁multi": 6079,
+ "▁gru": 6080,
+ "▁Jo": 6081,
+ "▁Canada": 6082,
+ "▁Thomas": 6083,
+ "▁larger": 6084,
+ "▁compar": 6085,
+ "Current": 6086,
+ "that": 6087,
+ "▁drop": 6088,
+ "ент": 6089,
+ "▁Republic": 6090,
+ "▁dise": 6091,
+ "▁effects": 6092,
+ "▁girls": 6093,
+ "encies": 6094,
+ "ellig": 6095,
+ "▁Note": 6096,
+ "▁Associ": 6097,
+ "▁uses": 6098,
+ "elled": 6099,
+ "▁warm": 6100,
+ "thread": 6101,
+ "font": 6102,
+ "▁zum": 6103,
+ "▁follows": 6104,
+ "▁whom": 6105,
+ "TA": 6106,
+ "▁wild": 6107,
+ "▁AR": 6108,
+ "iable": 6109,
+ "▁True": 6110,
+ "Position": 6111,
+ "▁sell": 6112,
+ "cher": 6113,
+ "▁Bus": 6114,
+ "▁lean": 6115,
+ "ACE": 6116,
+ "▁served": 6117,
+ "hw": 6118,
+ "▁Cur": 6119,
+ "▁north": 6120,
+ "Dat": 6121,
+ "▁>>": 6122,
+ "command": 6123,
+ "atz": 6124,
+ "▁mal": 6125,
+ "став": 6126,
+ "▁Press": 6127,
+ "▁characters": 6128,
+ "▁zero": 6129,
+ "AGE": 6130,
+ "rapper": 6131,
+ "▁kitchen": 6132,
+ "aming": 6133,
+ "▁restr": 6134,
+ "XX": 6135,
+ "▁College": 6136,
+ "▁Array": 6137,
+ "▁fresh": 6138,
+ "▁shift": 6139,
+ "▁specified": 6140,
+ "plete": 6141,
+ "ITE": 6142,
+ "▁Camp": 6143,
+ "rial": 6144,
+ "cb": 6145,
+ "▁TH": 6146,
+ "IB": 6147,
+ "osen": 6148,
+ "▁ú": 6149,
+ "▁params": 6150,
+ "ignment": 6151,
+ "adding": 6152,
+ "▁degree": 6153,
+ "Local": 6154,
+ "Oh": 6155,
+ "▁zur": 6156,
+ "▁levels": 6157,
+ "CS": 6158,
+ "finished": 6159,
+ "Case": 6160,
+ "riage": 6161,
+ "Vector": 6162,
+ "▁sea": 6163,
+ "antic": 6164,
+ "▁League": 6165,
+ "▁therefore": 6166,
+ "One": 6167,
+ "Return": 6168,
+ "Access": 6169,
+ "vas": 6170,
+ "▁ос": 6171,
+ "▁rat": 6172,
+ "Big": 6173,
+ "▁behavior": 6174,
+ "kr": 6175,
+ "▁undefined": 6176,
+ "▁Es": 6177,
+ "▁appeared": 6178,
+ "eles": 6179,
+ "▁WAR": 6180,
+ "Stat": 6181,
+ "▁Google": 6182,
+ "▁credit": 6183,
+ "▁File": 6184,
+ "anging": 6185,
+ "house": 6186,
+ "romise": 6187,
+ "gent": 6188,
+ "▁habit": 6189,
+ "▁society": 6190,
+ "▁encour": 6191,
+ "▁paint": 6192,
+ "pet": 6193,
+ "▁UK": 6194,
+ "aws": 6195,
+ "onom": 6196,
+ "Gl": 6197,
+ "}_{\\": 6198,
+ "eless": 6199,
+ "emy": 6200,
+ "▁Cong": 6201,
+ "▁developed": 6202,
+ "▁images": 6203,
+ "▁ö": 6204,
+ "▁font": 6205,
+ "clear": 6206,
+ "gin": 6207,
+ "▁Lord": 6208,
+ "▁transport": 6209,
+ "▁::": 6210,
+ "▁cup": 6211,
+ "ulate": 6212,
+ "▁During": 6213,
+ "priv": 6214,
+ "▁extrem": 6215,
+ "▁Di": 6216,
+ "▁doubt": 6217,
+ "Py": 6218,
+ "ifying": 6219,
+ "split": 6220,
+ "ego": 6221,
+ "github": 6222,
+ "▁),": 6223,
+ "ROM": 6224,
+ "▁chair": 6225,
+ "▁trade": 6226,
+ "▁nicht": 6227,
+ "Top": 6228,
+ "Store": 6229,
+ "▁parte": 6230,
+ "project": 6231,
+ "nia": 6232,
+ "▁від": 6233,
+ "war": 6234,
+ "▁Prof": 6235,
+ "▁caught": 6236,
+ "Thread": 6237,
+ "ства": 6238,
+ "author": 6239,
+ "▁doll": 6240,
+ "▁harm": 6241,
+ "▁Gen": 6242,
+ "tree": 6243,
+ "etime": 6244,
+ "cfg": 6245,
+ "▁guys": 6246,
+ "▁California": 6247,
+ "▁Green": 6248,
+ "▁movement": 6249,
+ "iej": 6250,
+ "▁statement": 6251,
+ "▁seeing": 6252,
+ "▁haven": 6253,
+ "vention": 6254,
+ "SL": 6255,
+ "chedul": 6256,
+ "iert": 6257,
+ "▁primary": 6258,
+ "▁civil": 6259,
+ "rian": 6260,
+ "▁button": 6261,
+ "▁lived": 6262,
+ "Pass": 6263,
+ "sor": 6264,
+ "▁watching": 6265,
+ "▁skills": 6266,
+ "tee": 6267,
+ "Level": 6268,
+ "▁scient": 6269,
+ "hs": 6270,
+ "▁agre": 6271,
+ "cat": 6272,
+ "▁tend": 6273,
+ "▁Mill": 6274,
+ "▁Cap": 6275,
+ "ORD": 6276,
+ "gle": 6277,
+ "▁сво": 6278,
+ "»,": 6279,
+ "▁ahead": 6280,
+ "vest": 6281,
+ "▁Jose": 6282,
+ "ischer": 6283,
+ "și": 6284,
+ "▁leaving": 6285,
+ "▁для": 6286,
+ "▁south": 6287,
+ "▁consum": 6288,
+ "Range": 6289,
+ "▁activities": 6290,
+ "Sec": 6291,
+ "▁sales": 6292,
+ "▁fix": 6293,
+ "▁jed": 6294,
+ "rum": 6295,
+ "vector": 6296,
+ "▁spot": 6297,
+ "▁manufact": 6298,
+ "кт": 6299,
+ "orrow": 6300,
+ "sign": 6301,
+ "▁college": 6302,
+ "▁driver": 6303,
+ "▁definitely": 6304,
+ "▁spend": 6305,
+ "mission": 6306,
+ "зу": 6307,
+ "atively": 6308,
+ "bi": 6309,
+ "Callback": 6310,
+ "▁particularly": 6311,
+ "▁hell": 6312,
+ "▁pool": 6313,
+ "PRE": 6314,
+ "▁clearly": 6315,
+ "PT": 6316,
+ "othes": 6317,
+ "▁Id": 6318,
+ "Location": 6319,
+ "▁Run": 6320,
+ "▁fixed": 6321,
+ "▁Hand": 6322,
+ "bal": 6323,
+ "double": 6324,
+ "Can": 6325,
+ "Omega": 6326,
+ "▁challeng": 6327,
+ "▁standing": 6328,
+ "iten": 6329,
+ "▁mechan": 6330,
+ "▁durch": 6331,
+ "▁dell": 6332,
+ "▁raised": 6333,
+ "▁weak": 6334,
+ "▁Du": 6335,
+ "grad": 6336,
+ "▁scene": 6337,
+ "poss": 6338,
+ "▁ton": 6339,
+ "▁earth": 6340,
+ "ulations": 6341,
+ "▁strength": 6342,
+ "aked": 6343,
+ "▁remain": 6344,
+ "▁Bi": 6345,
+ "▁customer": 6346,
+ "range": 6347,
+ "▁interested": 6348,
+ "ONE": 6349,
+ "▁coff": 6350,
+ "require": 6351,
+ "▁Only": 6352,
+ "▁Web": 6353,
+ "▁farm": 6354,
+ "▁activity": 6355,
+ "▁rout": 6356,
+ "bling": 6357,
+ "SY": 6358,
+ "▁Richard": 6359,
+ "▁Ref": 6360,
+ "▁кон": 6361,
+ "▁jun": 6362,
+ "born": 6363,
+ "ijn": 6364,
+ "Configuration": 6365,
+ "uman": 6366,
+ "EE": 6367,
+ "▁married": 6368,
+ "▁За": 6369,
+ "▁fat": 6370,
+ "▁kid": 6371,
+ "▁Tur": 6372,
+ "▁offered": 6373,
+ "nic": 6374,
+ "▁Big": 6375,
+ "Gamma": 6376,
+ "▁Health": 6377,
+ "▁TR": 6378,
+ "▁się": 6379,
+ "▁construction": 6380,
+ "▁Church": 6381,
+ "▁Bet": 6382,
+ "bus": 6383,
+ "▁earn": 6384,
+ "rict": 6385,
+ "▁пра": 6386,
+ "▁brain": 6387,
+ "▁fra": 6388,
+ "▁Op": 6389,
+ "FIG": 6390,
+ "ema": 6391,
+ "▁European": 6392,
+ "▁Saint": 6393,
+ "ARE": 6394,
+ "uri": 6395,
+ "▁River": 6396,
+ "{}": 6397,
+ "▁sitting": 6398,
+ "▁understanding": 6399,
+ "▁plans": 6400,
+ "ropri": 6401,
+ "▁older": 6402,
+ "▁pressure": 6403,
+ "Impl": 6404,
+ "▁peace": 6405,
+ "Connection": 6406,
+ "▁fi": 6407,
+ "rich": 6408,
+ "▁shut": 6409,
+ "apers": 6410,
+ "Port": 6411,
+ "▁Look": 6412,
+ "rim": 6413,
+ "auth": 6414,
+ "auto": 6415,
+ "▁highly": 6416,
+ "▁unless": 6417,
+ "▁Wal": 6418,
+ "▁ren": 6419,
+ "ws": 6420,
+ "▁core": 6421,
+ "(-": 6422,
+ "▁clim": 6423,
+ "ruit": 6424,
+ "▁callback": 6425,
+ "hest": 6426,
+ "▁Charles": 6427,
+ "▁Long": 6428,
+ "}=": 6429,
+ "ър": 6430,
+ "▁shared": 6431,
+ "ulated": 6432,
+ "gorithm": 6433,
+ "▁Home": 6434,
+ "▁village": 6435,
+ "ees": 6436,
+ "sv": 6437,
+ "▁restaur": 6438,
+ "rey": 6439,
+ "▁Cast": 6440,
+ "▁Person": 6441,
+ "кий": 6442,
+ "▁organiz": 6443,
+ "▁Rad": 6444,
+ "ponents": 6445,
+ "▁werden": 6446,
+ "▁bow": 6447,
+ "sen": 6448,
+ "ami": 6449,
+ "Interface": 6450,
+ "▁basis": 6451,
+ "▁Company": 6452,
+ "ernel": 6453,
+ "itu": 6454,
+ "Hash": 6455,
+ "▁aan": 6456,
+ "▁х": 6457,
+ "▁smile": 6458,
+ "xml": 6459,
+ "▁scen": 6460,
+ "amm": 6461,
+ "tool": 6462,
+ "aria": 6463,
+ "▁accur": 6464,
+ "settings": 6465,
+ "▁Jesus": 6466,
+ "acement": 6467,
+ "power": 6468,
+ "(!": 6469,
+ "▁calls": 6470,
+ "▁basic": 6471,
+ "▁settings": 6472,
+ "ript": 6473,
+ "pool": 6474,
+ "ctors": 6475,
+ "▁Foundation": 6476,
+ "▁weap": 6477,
+ "KEY": 6478,
+ "foot": 6479,
+ "▁radio": 6480,
+ "▁helped": 6481,
+ "mann": 6482,
+ "▁jump": 6483,
+ "▁tick": 6484,
+ "▁growing": 6485,
+ "aten": 6486,
+ "real": 6487,
+ "▁increasing": 6488,
+ "Device": 6489,
+ "varepsilon": 6490,
+ "▁sets": 6491,
+ "▁advant": 6492,
+ "Open": 6493,
+ "▁reasons": 6494,
+ "▁supposed": 6495,
+ "oes": 6496,
+ "ede": 6497,
+ "teen": 6498,
+ "ifdef": 6499,
+ "▁delete": 6500,
+ "▁&=": 6501,
+ "▁Bill": 6502,
+ "▁aim": 6503,
+ "▁Ok": 6504,
+ "▁Av": 6505,
+ "reci": 6506,
+ "acks": 6507,
+ "iste": 6508,
+ "Properties": 6509,
+ "▁tmp": 6510,
+ "▁dei": 6511,
+ "PER": 6512,
+ "DC": 6513,
+ "sta": 6514,
+ "нии": 6515,
+ "▁limited": 6516,
+ "▁greater": 6517,
+ "description": 6518,
+ "ori": 6519,
+ "aints": 6520,
+ "▁hy": 6521,
+ "▁Mel": 6522,
+ "▁CH": 6523,
+ "cons": 6524,
+ "▁surround": 6525,
+ "▁Who": 6526,
+ "arc": 6527,
+ "▁telev": 6528,
+ "itution": 6529,
+ "▁equal": 6530,
+ "кі": 6531,
+ "▁Israel": 6532,
+ "äh": 6533,
+ "▁Caption": 6534,
+ "▁exerc": 6535,
+ "empor": 6536,
+ "▁++": 6537,
+ "▁lib": 6538,
+ "make": 6539,
+ "▁MA": 6540,
+ "copy": 6541,
+ "friend": 6542,
+ "▁кото": 6543,
+ "▁damage": 6544,
+ "▁\\,": 6545,
+ "oded": 6546,
+ "▁none": 6547,
+ "▁evalu": 6548,
+ "ston": 6549,
+ ">,": 6550,
+ "FOR": 6551,
+ "▁norm": 6552,
+ "appe": 6553,
+ "Session": 6554,
+ "▁adult": 6555,
+ "▁hospital": 6556,
+ "▁recommend": 6557,
+ "property": 6558,
+ "stein": 6559,
+ "final": 6560,
+ "▁nu": 6561,
+ "second": 6562,
+ "▁aspect": 6563,
+ "\")]": 6564,
+ "жен": 6565,
+ "amento": 6566,
+ "▁rac": 6567,
+ "save": 6568,
+ "▁football": 6569,
+ "Ab": 6570,
+ "ungs": 6571,
+ "abil": 6572,
+ "▁Arch": 6573,
+ "system": 6574,
+ "hist": 6575,
+ "▁luck": 6576,
+ "render": 6577,
+ "▁sein": 6578,
+ "ioni": 6579,
+ "▁rot": 6580,
+ "▁corner": 6581,
+ "▁appropri": 6582,
+ "▁Software": 6583,
+ "▁tele": 6584,
+ "Delete": 6585,
+ "▁According": 6586,
+ "▁prison": 6587,
+ "▁lic": 6588,
+ "▁ми": 6589,
+ "term": 6590,
+ "sets": 6591,
+ "▁vel": 6592,
+ "▁rank": 6593,
+ "▁existing": 6594,
+ "▁Vir": 6595,
+ "▁trip": 6596,
+ "▁му": 6597,
+ "avax": 6598,
+ "▁ris": 6599,
+ "▁define": 6600,
+ "▁heat": 6601,
+ "car": 6602,
+ "▁convert": 6603,
+ "email": 6604,
+ "▁Under": 6605,
+ "▁Ш": 6606,
+ "▁Grand": 6607,
+ "▁exists": 6608,
+ "sys": 6609,
+ "eff": 6610,
+ "▁Top": 6611,
+ "▁č": 6612,
+ "▁tempor": 6613,
+ "▁arguments": 6614,
+ "▁supported": 6615,
+ "ensed": 6616,
+ "▁Francis": 6617,
+ "▁coord": 6618,
+ "▁achieve": 6619,
+ "▁Name": 6620,
+ "▁Jahr": 6621,
+ "▁Gi": 6622,
+ "she": 6623,
+ "▁Dev": 6624,
+ "▁alla": 6625,
+ "▁WIT": 6626,
+ "agment": 6627,
+ "custom": 6628,
+ "alls": 6629,
+ "&&": 6630,
+ "WE": 6631,
+ "▁holding": 6632,
+ "prototype": 6633,
+ "▁fing": 6634,
+ "▁bag": 6635,
+ "▁Party": 6636,
+ "stack": 6637,
+ "▁economic": 6638,
+ "▁Gal": 6639,
+ "idents": 6640,
+ "▁Jun": 6641,
+ "▁showed": 6642,
+ "osh": 6643,
+ "▁Bay": 6644,
+ "mail": 6645,
+ "▁SO": 6646,
+ "▁\"<": 6647,
+ "graphics": 6648,
+ "▁fu": 6649,
+ "click": 6650,
+ "▁battle": 6651,
+ "{{": 6652,
+ "▁Event": 6653,
+ "rior": 6654,
+ "chaft": 6655,
+ "▁favorite": 6656,
+ "usive": 6657,
+ "support": 6658,
+ "bm": 6659,
+ "Kind": 6660,
+ "▁safety": 6661,
+ "▁Ent": 6662,
+ "cup": 6663,
+ "▁Australia": 6664,
+ "▁destroy": 6665,
+ "▁organization": 6666,
+ "iden": 6667,
+ "################": 6668,
+ "dec": 6669,
+ "▁za": 6670,
+ "▁seven": 6671,
+ "arely": 6672,
+ "▁flag": 6673,
+ "Dir": 6674,
+ "▁Carl": 6675,
+ "▁doctor": 6676,
+ "▁variety": 6677,
+ "▁Lin": 6678,
+ "▁tom": 6679,
+ "^{(": 6680,
+ "Bo": 6681,
+ "antes": 6682,
+ "▁mine": 6683,
+ "▁Mit": 6684,
+ "▁describe": 6685,
+ "Args": 6686,
+ "LS": 6687,
+ "API": 6688,
+ "▁Luc": 6689,
+ "phone": 6690,
+ "▁science": 6691,
+ "▁Oper": 6692,
+ "Next": 6693,
+ "▁investig": 6694,
+ "▁demonstr": 6695,
+ "▁Govern": 6696,
+ "▁objects": 6697,
+ "▁Louis": 6698,
+ "▁Returns": 6699,
+ "▁han": 6700,
+ "nam": 6701,
+ "▁comme": 6702,
+ "▁presence": 6703,
+ "▁pel": 6704,
+ "▁detect": 6705,
+ ")=": 6706,
+ "▁Chinese": 6707,
+ "▁rich": 6708,
+ "▁classes": 6709,
+ "▁expand": 6710,
+ "▁Dom": 6711,
+ "▁Dec": 6712,
+ "sn": 6713,
+ "peed": 6714,
+ "▁Jim": 6715,
+ "should": 6716,
+ "▁Smith": 6717,
+ "▁pages": 6718,
+ "▁Jean": 6719,
+ "rics": 6720,
+ "▁Sund": 6721,
+ "ads": 6722,
+ "▁Their": 6723,
+ "unicip": 6724,
+ "ву": 6725,
+ "▁download": 6726,
+ "▁stress": 6727,
+ "▁Pet": 6728,
+ "menu": 6729,
+ "reme": 6730,
+ "▁compared": 6731,
+ "Ste": 6732,
+ "IND": 6733,
+ "container": 6734,
+ "▁Indian": 6735,
+ "oren": 6736,
+ "▁ses": 6737,
+ "▁Whe": 6738,
+ "▁roku": 6739,
+ "▁established": 6740,
+ "▁generally": 6741,
+ "▁fle": 6742,
+ "__(": 6743,
+ "=\"+": 6744,
+ "Var": 6745,
+ "▁Make": 6746,
+ "▁removed": 6747,
+ "zz": 6748,
+ "ün": 6749,
+ "▁mix": 6750,
+ "erk": 6751,
+ "iation": 6752,
+ "outer": 6753,
+ "SK": 6754,
+ "▁becomes": 6755,
+ "▁Hall": 6756,
+ "scious": 6757,
+ "▁watched": 6758,
+ "▁gather": 6759,
+ "▁Result": 6760,
+ "proof": 6761,
+ "pay": 6762,
+ "▁produced": 6763,
+ "▁|=": 6764,
+ "▁border": 6765,
+ "▁din": 6766,
+ "▁script": 6767,
+ "▁actions": 6768,
+ "▁mas": 6769,
+ "ща": 6770,
+ "ooth": 6771,
+ "▁Techn": 6772,
+ "Json": 6773,
+ "▁filled": 6774,
+ "ден": 6775,
+ "undle": 6776,
+ "сту": 6777,
+ "Tool": 6778,
+ "▁king": 6779,
+ "▁ven": 6780,
+ "stra": 6781,
+ "▁predict": 6782,
+ "▁lui": 6783,
+ "▁WARRAN": 6784,
+ "▁Fun": 6785,
+ "Script": 6786,
+ "▁powerful": 6787,
+ "▁lose": 6788,
+ "atically": 6789,
+ "▁daily": 6790,
+ "▁ring": 6791,
+ "▁arrived": 6792,
+ "Stack": 6793,
+ "scope": 6794,
+ "▁Back": 6795,
+ "elij": 6796,
+ "▁ze": 6797,
+ "keys": 6798,
+ "{\"": 6799,
+ "VID": 6800,
+ "▁license": 6801,
+ "what": 6802,
+ "▁proced": 6803,
+ "rant": 6804,
+ "estival": 6805,
+ "agram": 6806,
+ "▁LO": 6807,
+ "▁Henry": 6808,
+ "▁flags": 6809,
+ "Down": 6810,
+ "scription": 6811,
+ "▁families": 6812,
+ "isse": 6813,
+ "bour": 6814,
+ "▁Bur": 6815,
+ "—\"": 6816,
+ "▁brief": 6817,
+ "▁creating": 6818,
+ "▁clients": 6819,
+ "rangle": 6820,
+ "▁amazing": 6821,
+ "▁sind": 6822,
+ "▁covered": 6823,
+ "Well": 6824,
+ "сте": 6825,
+ "тор": 6826,
+ "▁Bas": 6827,
+ "total": 6828,
+ "▁Init": 6829,
+ "▁sand": 6830,
+ "Unit": 6831,
+ "▁murder": 6832,
+ "▁bright": 6833,
+ "▁trav": 6834,
+ "icans": 6835,
+ "▁attribute": 6836,
+ "fc": 6837,
+ "▁placed": 6838,
+ "EST": 6839,
+ "Vari": 6840,
+ "▁cos": 6841,
+ "▁attract": 6842,
+ "anel": 6843,
+ "}).": 6844,
+ "bytes": 6845,
+ "▁parse": 6846,
+ "▁belong": 6847,
+ "BN": 6848,
+ "▁Sol": 6849,
+ "Po": 6850,
+ "`,": 6851,
+ "▁calling": 6852,
+ "▁?>": 6853,
+ "▁iter": 6854,
+ "▁url": 6855,
+ "▁evening": 6856,
+ "reek": 6857,
+ "▁honest": 6858,
+ "▁director": 6859,
+ "RC": 6860,
+ "▁solid": 6861,
+ "▁phil": 6862,
+ "iene": 6863,
+ "FAULT": 6864,
+ "cope": 6865,
+ "▁History": 6866,
+ "▁Team": 6867,
+ "reedom": 6868,
+ "▁ru": 6869,
+ "UB": 6870,
+ "▁worse": 6871,
+ "imo": 6872,
+ "Mat": 6873,
+ "▁Mex": 6874,
+ "actor": 6875,
+ "▁vor": 6876,
+ "ться": 6877,
+ "▁experiment": 6878,
+ "▁Play": 6879,
+ "▁Another": 6880,
+ "▁happens": 6881,
+ "uan": 6882,
+ "▁patients": 6883,
+ "▁rend": 6884,
+ "▁Mo": 6885,
+ "▁Tex": 6886,
+ "▁wed": 6887,
+ "tn": 6888,
+ "insert": 6889,
+ "▁па": 6890,
+ "▁anti": 6891,
+ "Match": 6892,
+ "ampionship": 6893,
+ "▁forces": 6894,
+ "▁Hot": 6895,
+ "▁phase": 6896,
+ "▁template": 6897,
+ "stop": 6898,
+ "icated": 6899,
+ "▁managed": 6900,
+ "wait": 6901,
+ "▁*(": 6902,
+ "GB": 6903,
+ "▁appoint": 6904,
+ "ła": 6905,
+ "▁stick": 6906,
+ "▁FOR": 6907,
+ "▁Vis": 6908,
+ "tor": 6909,
+ "▁př": 6910,
+ "quest": 6911,
+ "uses": 6912,
+ "\");\r": 6913,
+ "▁suddenly": 6914,
+ "éc": 6915,
+ "ND": 6916,
+ "urop": 6917,
+ "ред": 6918,
+ "▁insurance": 6919,
+ "access": 6920,
+ "unfinished": 6921,
+ "▁tamb": 6922,
+ "▁sac": 6923,
+ "▁Court": 6924,
+ "▁missing": 6925,
+ "▁Where": 6926,
+ "▁Sum": 6927,
+ "}^{\\": 6928,
+ "▁sua": 6929,
+ "_,": 6930,
+ "▁thick": 6931,
+ "▁Trump": 6932,
+ "▁operations": 6933,
+ "FS": 6934,
+ "▁deux": 6935,
+ "dz": 6936,
+ "Template": 6937,
+ "▁\"/": 6938,
+ "▁odd": 6939,
+ "▁reality": 6940,
+ "▁teams": 6941,
+ "▁cer": 6942,
+ "oma": 6943,
+ "▁și": 6944,
+ "▁cloud": 6945,
+ "▁Department": 6946,
+ "Ne": 6947,
+ "▁requires": 6948,
+ "items": 6949,
+ "▁III": 6950,
+ "rightarrow": 6951,
+ ")->": 6952,
+ "▁writer": 6953,
+ "replace": 6954,
+ "▁thr": 6955,
+ "jen": 6956,
+ "▁ot": 6957,
+ "▁occup": 6958,
+ "▁eventually": 6959,
+ "▁Math": 6960,
+ "▁conserv": 6961,
+ "amer": 6962,
+ "▁Fort": 6963,
+ "▁dry": 6964,
+ "▁sexual": 6965,
+ "▁costs": 6966,
+ "▁forms": 6967,
+ "▁Vict": 6968,
+ "PAR": 6969,
+ "framework": 6970,
+ "▁ди": 6971,
+ "Operation": 6972,
+ "зна": 6973,
+ "which": 6974,
+ "▁tight": 6975,
+ "Invalid": 6976,
+ "▁partner": 6977,
+ "▁пред": 6978,
+ "▁thank": 6979,
+ "▁guard": 6980,
+ "hem": 6981,
+ "Body": 6982,
+ "▁emot": 6983,
+ "IX": 6984,
+ "fast": 6985,
+ "що": 6986,
+ "ño": 6987,
+ "night": 6988,
+ "▁Sci": 6989,
+ "ника": 6990,
+ "▁TO": 6991,
+ "▁individuals": 6992,
+ "сси": 6993,
+ "}),": 6994,
+ "False": 6995,
+ "(\"%": 6996,
+ "▁optim": 6997,
+ "▁-->": 6998,
+ "▁factor": 6999,
+ "▁smaller": 7000,
+ "▁contain": 7001,
+ "spect": 7002,
+ "Engine": 7003,
+ "▁announced": 7004,
+ "▁Democr": 7005,
+ "▁rob": 7006,
+ "▁flat": 7007,
+ "osoph": 7008,
+ "Search": 7009,
+ "ahl": 7010,
+ "▁Exception": 7011,
+ "▁Ol": 7012,
+ "equals": 7013,
+ "▁unter": 7014,
+ "shape": 7015,
+ "NS": 7016,
+ "Obj": 7017,
+ "▁species": 7018,
+ "weight": 7019,
+ "you": 7020,
+ "▁este": 7021,
+ "▁View": 7022,
+ "▁mission": 7023,
+ "▁journal": 7024,
+ "Values": 7025,
+ "▁einem": 7026,
+ "ismo": 7027,
+ "▁projects": 7028,
+ "▁Das": 7029,
+ "rible": 7030,
+ "▁serve": 7031,
+ "▁opening": 7032,
+ "▁hur": 7033,
+ "▁programs": 7034,
+ "▁USA": 7035,
+ "iliar": 7036,
+ "idos": 7037,
+ "Br": 7038,
+ "estamp": 7039,
+ "▁tools": 7040,
+ "anner": 7041,
+ "RT": 7042,
+ "▁Start": 7043,
+ "▁bath": 7044,
+ "▁coffee": 7045,
+ "orter": 7046,
+ "internal": 7047,
+ "files": 7048,
+ "INVAL": 7049,
+ "ako": 7050,
+ "dt": 7051,
+ "▁Second": 7052,
+ "▁alloc": 7053,
+ "▁ended": 7054,
+ "acional": 7055,
+ "▁manager": 7056,
+ "▁Sun": 7057,
+ "agg": 7058,
+ "▁leader": 7059,
+ "olved": 7060,
+ "▁что": 7061,
+ "▁traditional": 7062,
+ "shot": 7063,
+ "rup": 7064,
+ "CF": 7065,
+ "▁Each": 7066,
+ "wr": 7067,
+ "▁Som": 7068,
+ "▁materials": 7069,
+ "▁msg": 7070,
+ "▁syn": 7071,
+ "▁produce": 7072,
+ "▁storage": 7073,
+ "subsection": 7074,
+ "▁Sie": 7075,
+ "▁IP": 7076,
+ "CESS": 7077,
+ "▁wa": 7078,
+ "Record": 7079,
+ "▁marketing": 7080,
+ "plet": 7081,
+ "Dialog": 7082,
+ "▁mentioned": 7083,
+ "▁Na": 7084,
+ "▁Union": 7085,
+ "▁API": 7086,
+ "▁negative": 7087,
+ "txt": 7088,
+ "▁easier": 7089,
+ "legal": 7090,
+ "Dep": 7091,
+ "▁novel": 7092,
+ "eur": 7093,
+ "ació": 7094,
+ "▁Bud": 7095,
+ "▁carry": 7096,
+ "schaft": 7097,
+ "▁broken": 7098,
+ "▁trees": 7099,
+ ">();": 7100,
+ "▁emb": 7101,
+ "ieder": 7102,
+ "▁route": 7103,
+ "ikel": 7104,
+ "▁listen": 7105,
+ "ashion": 7106,
+ "▁Mrs": 7107,
+ "▁equipment": 7108,
+ "agger": 7109,
+ "▁Thus": 7110,
+ "▁matrix": 7111,
+ "alla": 7112,
+ "▁Tour": 7113,
+ "▁conversation": 7114,
+ "Mon": 7115,
+ "ournal": 7116,
+ "▁minute": 7117,
+ "Am": 7118,
+ "Api": 7119,
+ "▁forget": 7120,
+ "Me": 7121,
+ "levant": 7122,
+ "temp": 7123,
+ "▁telling": 7124,
+ "move": 7125,
+ "▁independent": 7126,
+ "toString": 7127,
+ "edit": 7128,
+ "▁Jac": 7129,
+ "azz": 7130,
+ "react": 7131,
+ "▁cin": 7132,
+ "▁Prov": 7133,
+ "isted": 7134,
+ "▁hash": 7135,
+ "onna": 7136,
+ "iki": 7137,
+ "▁generated": 7138,
+ "Render": 7139,
+ "▁psych": 7140,
+ "nav": 7141,
+ "▁entr": 7142,
+ "пра": 7143,
+ "rx": 7144,
+ "ATH": 7145,
+ "▁assume": 7146,
+ "Tree": 7147,
+ "sembly": 7148,
+ "▁Matt": 7149,
+ "caption": 7150,
+ "▁solutions": 7151,
+ "▁faith": 7152,
+ "▁digital": 7153,
+ "▁excell": 7154,
+ "▁Version": 7155,
+ "Debug": 7156,
+ "▁жи": 7157,
+ "▁carried": 7158,
+ "reset": 7159,
+ "▁slowly": 7160,
+ "ancing": 7161,
+ "▁owner": 7162,
+ "▁Ter": 7163,
+ "▁Did": 7164,
+ "▁gest": 7165,
+ "▁été": 7166,
+ "▁proof": 7167,
+ "Font": 7168,
+ "▁nob": 7169,
+ "Co": 7170,
+ "▁GNU": 7171,
+ "▁liber": 7172,
+ "itness": 7173,
+ "▁hij": 7174,
+ "▁vert": 7175,
+ "ша": 7176,
+ "FLAG": 7177,
+ "MENT": 7178,
+ "▁Son": 7179,
+ "Mult": 7180,
+ "▁district": 7181,
+ "connect": 7182,
+ "jection": 7183,
+ "lymp": 7184,
+ "▁realized": 7185,
+ "mos": 7186,
+ "ye": 7187,
+ "▁render": 7188,
+ "rio": 7189,
+ "▁interpret": 7190,
+ "▁slightly": 7191,
+ "fix": 7192,
+ "▁studies": 7193,
+ "▁rid": 7194,
+ "atre": 7195,
+ "▁benefits": 7196,
+ "▁Face": 7197,
+ "ivery": 7198,
+ "рия": 7199,
+ "document": 7200,
+ "▁asking": 7201,
+ "Last": 7202,
+ "arante": 7203,
+ "▁Martin": 7204,
+ "▁Ell": 7205,
+ "▁vector": 7206,
+ "▁forced": 7207,
+ "оло": 7208,
+ "PH": 7209,
+ "WR": 7210,
+ "▁Kl": 7211,
+ "▁sky": 7212,
+ "▁strategy": 7213,
+ "ocked": 7214,
+ "▁neck": 7215,
+ "ści": 7216,
+ "OUT": 7217,
+ ")),": 7218,
+ "Custom": 7219,
+ "▁wie": 7220,
+ "▁sweet": 7221,
+ "▁temp": 7222,
+ "▁foreign": 7223,
+ "▁hall": 7224,
+ "astr": 7225,
+ "Ass": 7226,
+ "MODE": 7227,
+ "▁maximum": 7228,
+ "annels": 7229,
+ "▁tip": 7230,
+ "▁seconds": 7231,
+ "▁stack": 7232,
+ "iga": 7233,
+ "▁raise": 7234,
+ "enable": 7235,
+ "oir": 7236,
+ "▁soul": 7237,
+ "Ke": 7238,
+ ")$.": 7239,
+ "▁Tim": 7240,
+ "ALSE": 7241,
+ "iser": 7242,
+ "contin": 7243,
+ "bel": 7244,
+ "▁mad": 7245,
+ "lichen": 7246,
+ "abe": 7247,
+ "safe": 7248,
+ "▁concent": 7249,
+ "bound": 7250,
+ "▁Requ": 7251,
+ "switch": 7252,
+ "▁stone": 7253,
+ "▁transl": 7254,
+ "▁vac": 7255,
+ "andon": 7256,
+ "▁Fore": 7257,
+ "▁sounds": 7258,
+ "▁Pop": 7259,
+ "▁HT": 7260,
+ "lia": 7261,
+ "enter": 7262,
+ "▁helps": 7263,
+ "edy": 7264,
+ "ствен": 7265,
+ "anted": 7266,
+ "▁Its": 7267,
+ "▁Step": 7268,
+ "Icon": 7269,
+ "▁EXPECT": 7270,
+ "ialized": 7271,
+ "Post": 7272,
+ "aze": 7273,
+ "▁Carol": 7274,
+ "▁req": 7275,
+ "▁critical": 7276,
+ "DS": 7277,
+ "▁seat": 7278,
+ "aped": 7279,
+ "▁upper": 7280,
+ "▁Sy": 7281,
+ "▁explain": 7282,
+ "▁'./": 7283,
+ "utils": 7284,
+ "possible": 7285,
+ "▁dont": 7286,
+ "Host": 7287,
+ "▁approxim": 7288,
+ "Async": 7289,
+ "▁grab": 7290,
+ "▁sources": 7291,
+ "▁Mos": 7292,
+ "▁Germany": 7293,
+ "▁rub": 7294,
+ "CHAN": 7295,
+ "▁rain": 7296,
+ "▁truly": 7297,
+ "▁joined": 7298,
+ "▁": 7299,
+ "▁Lo": 7300,
+ "Description": 7301,
+ "akt": 7302,
+ "▁Ann": 7303,
+ "^*": 7304,
+ "idae": 7305,
+ "(:": 7306,
+ "tw": 7307,
+ "Mar": 7308,
+ "produ": 7309,
+ "▁spoke": 7310,
+ "ют": 7311,
+ "▁walking": 7312,
+ "▁nodded": 7313,
+ "Props": 7314,
+ "Enabled": 7315,
+ "irk": 7316,
+ "FILE": 7317,
+ "equal": 7318,
+ "pping": 7319,
+ "oli": 7320,
+ "EV": 7321,
+ "enz": 7322,
+ "eting": 7323,
+ "▁sample": 7324,
+ "▁artist": 7325,
+ "[$": 7326,
+ "ità": 7327,
+ "йо": 7328,
+ "props": 7329,
+ "bu": 7330,
+ "ев": 7331,
+ "▁responsible": 7332,
+ "MT": 7333,
+ "▁caused": 7334,
+ "▁theme": 7335,
+ "▁Was": 7336,
+ "▁Before": 7337,
+ "acle": 7338,
+ "▁року": 7339,
+ "cu": 7340,
+ "DEV": 7341,
+ "▁hung": 7342,
+ "textbf": 7343,
+ "▁spin": 7344,
+ "▁latest": 7345,
+ "entially": 7346,
+ "▁Program": 7347,
+ "Metadata": 7348,
+ "password": 7349,
+ "▁hurt": 7350,
+ "кс": 7351,
+ "▁Aus": 7352,
+ "sey": 7353,
+ "allet": 7354,
+ "xF": 7355,
+ "▁Road": 7356,
+ "ется": 7357,
+ "▁rent": 7358,
+ "ция": 7359,
+ "▁Assert": 7360,
+ "іль": 7361,
+ "ück": 7362,
+ "▁sites": 7363,
+ "Document": 7364,
+ "▁obtained": 7365,
+ "▁ci": 7366,
+ "▁[\"": 7367,
+ "▁completed": 7368,
+ "aset": 7369,
+ "raid": 7370,
+ "▁sorry": 7371,
+ "▁fab": 7372,
+ "▁schools": 7373,
+ "ходи": 7374,
+ "▁scr": 7375,
+ "▁incor": 7376,
+ "▁'/": 7377,
+ "▁spr": 7378,
+ "▁Text": 7379,
+ "▁commercial": 7380,
+ "ingly": 7381,
+ "▁opinion": 7382,
+ "▁Star": 7383,
+ "Sign": 7384,
+ "▁javax": 7385,
+ "wi": 7386,
+ "lat": 7387,
+ "▁Key": 7388,
+ "varphi": 7389,
+ "ды": 7390,
+ "▁connected": 7391,
+ "▁adjust": 7392,
+ "▁Az": 7393,
+ "▁planning": 7394,
+ "---": 7395,
+ "Integer": 7396,
+ "auf": 7397,
+ "expected": 7398,
+ "▁fant": 7399,
+ "▁tou": 7400,
+ "Parent": 7401,
+ "▁Lat": 7402,
+ "▁thoughts": 7403,
+ "▁Jud": 7404,
+ "Parameters": 7405,
+ "Gr": 7406,
+ "ром": 7407,
+ "IA": 7408,
+ "▁Bob": 7409,
+ "lict": 7410,
+ "lan": 7411,
+ "omic": 7412,
+ "▁apart": 7413,
+ "▁trou": 7414,
+ "▁appreci": 7415,
+ "▁Christmas": 7416,
+ "irq": 7417,
+ "thon": 7418,
+ "▁Error": 7419,
+ "▁score": 7420,
+ "rome": 7421,
+ "▁neighbor": 7422,
+ "▁Mur": 7423,
+ "admin": 7424,
+ "▁Film": 7425,
+ "Rect": 7426,
+ "▁configuration": 7427,
+ "▁cs": 7428,
+ "gun": 7429,
+ "channel": 7430,
+ "▁Report": 7431,
+ "▁strateg": 7432,
+ "▁workers": 7433,
+ "fields": 7434,
+ "Schema": 7435,
+ "appa": 7436,
+ "olic": 7437,
+ "EO": 7438,
+ "▁Charl": 7439,
+ "▁Cup": 7440,
+ "png": 7441,
+ "▁Hill": 7442,
+ "owe": 7443,
+ "▁mostly": 7444,
+ "”.": 7445,
+ "▁finish": 7446,
+ "▁Со": 7447,
+ "▁stars": 7448,
+ "player": 7449,
+ "▁inner": 7450,
+ "component": 7451,
+ "tim": 7452,
+ "IE": 7453,
+ "▁ther": 7454,
+ "▁smart": 7455,
+ "▁sad": 7456,
+ "▁Council": 7457,
+ "area": 7458,
+ "lay": 7459,
+ "▁ба": 7460,
+ "▁gradu": 7461,
+ "▁chem": 7462,
+ "▁ho": 7463,
+ "Select": 7464,
+ "▁instr": 7465,
+ "▁kl": 7466,
+ "ifications": 7467,
+ "Long": 7468,
+ "▁sobre": 7469,
+ "▁Old": 7470,
+ "west": 7471,
+ "},\\": 7472,
+ "ingu": 7473,
+ "▁spring": 7474,
+ "▁nur": 7475,
+ "example": 7476,
+ "When": 7477,
+ "▁advice": 7478,
+ "▁ult": 7479,
+ "ennis": 7480,
+ "▁Love": 7481,
+ "▁\"\"": 7482,
+ "▁increased": 7483,
+ "▁finding": 7484,
+ "irty": 7485,
+ "istrict": 7486,
+ "▁layer": 7487,
+ "template": 7488,
+ "First": 7489,
+ "ным": 7490,
+ "igration": 7491,
+ "rency": 7492,
+ "owie": 7493,
+ "▁np": 7494,
+ "▁selection": 7495,
+ "▁Nach": 7496,
+ "▁PRO": 7497,
+ "▁polic": 7498,
+ "▁database": 7499,
+ "▁byte": 7500,
+ "▁providing": 7501,
+ "mac": 7502,
+ "▁metal": 7503,
+ "modules": 7504,
+ "▁Georg": 7505,
+ "▁Sa": 7506,
+ "▁establish": 7507,
+ "...\"": 7508,
+ "iu": 7509,
+ "kin": 7510,
+ "▁eth": 7511,
+ "▁Sand": 7512,
+ "▁Chapter": 7513,
+ "▁gal": 7514,
+ "▁ice": 7515,
+ "Red": 7516,
+ "▁dal": 7517,
+ "▁principal": 7518,
+ "Msg": 7519,
+ "▁remains": 7520,
+ "нг": 7521,
+ "Title": 7522,
+ "Rel": 7523,
+ "Display": 7524,
+ "Non": 7525,
+ "▁definition": 7526,
+ "▁attr": 7527,
+ "▁signal": 7528,
+ "hl": 7529,
+ "▁sel": 7530,
+ "▁volume": 7531,
+ "▁cache": 7532,
+ "hens": 7533,
+ "▁wird": 7534,
+ "[\\": 7535,
+ "NOT": 7536,
+ "▁election": 7537,
+ "utt": 7538,
+ "▁Window": 7539,
+ "ental": 7540,
+ "ifest": 7541,
+ "xf": 7542,
+ "▁Ра": 7543,
+ "▁overall": 7544,
+ "blic": 7545,
+ "▁editor": 7546,
+ "aden": 7547,
+ "▁cart": 7548,
+ "Left": 7549,
+ "uls": 7550,
+ "bing": 7551,
+ "Right": 7552,
+ "▁sé": 7553,
+ "Sim": 7554,
+ "▁camera": 7555,
+ "▁fav": 7556,
+ "Decl": 7557,
+ "spring": 7558,
+ "▁errors": 7559,
+ "Tab": 7560,
+ "println": 7561,
+ "▁Bern": 7562,
+ "nab": 7563,
+ "▁Base": 7564,
+ "▁auth": 7565,
+ "▁apparent": 7566,
+ "▁presented": 7567,
+ "▁remained": 7568,
+ "▁wet": 7569,
+ "Enc": 7570,
+ "INFO": 7571,
+ "▁Sing": 7572,
+ "package": 7573,
+ ")));": 7574,
+ "▁Social": 7575,
+ "▁Mass": 7576,
+ "▁despite": 7577,
+ "▁mobile": 7578,
+ "▁labor": 7579,
+ "Go": 7580,
+ "▁esp": 7581,
+ "▁Table": 7582,
+ "▁expert": 7583,
+ "▁flex": 7584,
+ "▁profession": 7585,
+ "▁pil": 7586,
+ "Collection": 7587,
+ "LOCK": 7588,
+ "▁applied": 7589,
+ "aller": 7590,
+ "orph": 7591,
+ "ENSE": 7592,
+ "▁был": 7593,
+ "▁db": 7594,
+ "overline": 7595,
+ "▁Code": 7596,
+ "▁bytes": 7597,
+ "▁trouble": 7598,
+ "▁насе": 7599,
+ "DD": 7600,
+ "▁Year": 7601,
+ "mbox": 7602,
+ "▁keeping": 7603,
+ "▁kick": 7604,
+ "äng": 7605,
+ "▁corresponding": 7606,
+ "▁library": 7607,
+ "▁*/\r": 7608,
+ "callback": 7609,
+ "ums": 7610,
+ "▁json": 7611,
+ "▁Mount": 7612,
+ "▁Stand": 7613,
+ "IGHT": 7614,
+ "▁News": 7615,
+ "▁comments": 7616,
+ "returns": 7617,
+ "Cal": 7618,
+ "▁award": 7619,
+ "▁bought": 7620,
+ "includegraphics": 7621,
+ "▁ле": 7622,
+ "dot": 7623,
+ "ronic": 7624,
+ "▁extremely": 7625,
+ "▁minor": 7626,
+ "ifer": 7627,
+ "java": 7628,
+ "endar": 7629,
+ "layout": 7630,
+ "plies": 7631,
+ "▁buf": 7632,
+ "▁Island": 7633,
+ "▁About": 7634,
+ "▁west": 7635,
+ "▁Scott": 7636,
+ "ACT": 7637,
+ "Why": 7638,
+ "▁largest": 7639,
+ "▁container": 7640,
+ "▁temperature": 7641,
+ "▁£": 7642,
+ "▁reduce": 7643,
+ "▁foi": 7644,
+ "han": 7645,
+ "▁bod": 7646,
+ "▁Van": 7647,
+ "▁nullptr": 7648,
+ "▁dating": 7649,
+ "▁chain": 7650,
+ "Flags": 7651,
+ "iento": 7652,
+ "sort": 7653,
+ "▁fan": 7654,
+ "▁determine": 7655,
+ "▁wear": 7656,
+ "BE": 7657,
+ "▁appropriate": 7658,
+ "лся": 7659,
+ "тов": 7660,
+ "▁goals": 7661,
+ "▁Map": 7662,
+ "▁Sar": 7663,
+ "▁Option": 7664,
+ "▁hate": 7665,
+ "▁zijn": 7666,
+ ",-": 7667,
+ "▁implied": 7668,
+ "bits": 7669,
+ "▁Men": 7670,
+ "skip": 7671,
+ "▁Mond": 7672,
+ "▁Hon": 7673,
+ "▁prove": 7674,
+ "van": 7675,
+ "▁traff": 7676,
+ "▁intr": 7677,
+ "pic": 7678,
+ "▁dropped": 7679,
+ "▁werd": 7680,
+ "▁separate": 7681,
+ "isa": 7682,
+ "▁tab": 7683,
+ "tml": 7684,
+ "▁\"$": 7685,
+ "mutex": 7686,
+ "▁Pan": 7687,
+ "serve": 7688,
+ "▁hotel": 7689,
+ "▁Last": 7690,
+ "step": 7691,
+ "▁vir": 7692,
+ "Rule": 7693,
+ "istan": 7694,
+ "oting": 7695,
+ "arks": 7696,
+ "(__": 7697,
+ "▁els": 7698,
+ "Player": 7699,
+ "]]": 7700,
+ "вич": 7701,
+ "ych": 7702,
+ "exception": 7703,
+ "=\"../": 7704,
+ "▁imagine": 7705,
+ "\"},": 7706,
+ "icago": 7707,
+ "eler": 7708,
+ "▁vs": 7709,
+ "▁Africa": 7710,
+ "▁Business": 7711,
+ "ocks": 7712,
+ "▁prz": 7713,
+ "▁fucking": 7714,
+ "▁picked": 7715,
+ "▁ві": 7716,
+ "▁\",": 7717,
+ "▁bott": 7718,
+ "▁failure": 7719,
+ "[:": 7720,
+ "▁Gar": 7721,
+ "apes": 7722,
+ "uple": 7723,
+ "▁fer": 7724,
+ "▁purchase": 7725,
+ "▁пер": 7726,
+ "▁bird": 7727,
+ "Widget": 7728,
+ "▁Sunday": 7729,
+ "▁Amaz": 7730,
+ "▁consult": 7731,
+ "utsch": 7732,
+ "anto": 7733,
+ "Storage": 7734,
+ "▁header": 7735,
+ "ühr": 7736,
+ "▁Ha": 7737,
+ "▁Association": 7738,
+ "▁sight": 7739,
+ "Cell": 7740,
+ "▁profile": 7741,
+ "▁female": 7742,
+ "ån": 7743,
+ "▁wid": 7744,
+ "zn": 7745,
+ "Direct": 7746,
+ "▁stret": 7747,
+ "aat": 7748,
+ "▁patient": 7749,
+ "here": 7750,
+ "▁Atl": 7751,
+ "inet": 7752,
+ "Definition": 7753,
+ "imary": 7754,
+ "Policy": 7755,
+ "▁dut": 7756,
+ "▁majority": 7757,
+ "сі": 7758,
+ "▁Project": 7759,
+ "ById": 7760,
+ "▁believed": 7761,
+ "▁Music": 7762,
+ "зы": 7763,
+ "anti": 7764,
+ "▁oder": 7765,
+ "Channel": 7766,
+ "▁sle": 7767,
+ "▁sequence": 7768,
+ "▁pieces": 7769,
+ "▁kne": 7770,
+ "▁absolutely": 7771,
+ "▁Philip": 7772,
+ "abilities": 7773,
+ "Que": 7774,
+ "▁Kar": 7775,
+ "Execut": 7776,
+ "▁Devel": 7777,
+ "▁electric": 7778,
+ "full": 7779,
+ "rolled": 7780,
+ "Dom": 7781,
+ "▁river": 7782,
+ "▁healthy": 7783,
+ "▁extern": 7784,
+ "fit": 7785,
+ "▁coach": 7786,
+ "▁Kr": 7787,
+ "asta": 7788,
+ "Compat": 7789,
+ "▁exit": 7790,
+ "▁Const": 7791,
+ "after": 7792,
+ "▁shoulder": 7793,
+ "▁jobs": 7794,
+ "zone": 7795,
+ "▁sale": 7796,
+ "ixel": 7797,
+ "▁determined": 7798,
+ "▁anyway": 7799,
+ "orf": 7800,
+ "▁Ger": 7801,
+ "allel": 7802,
+ "rees": 7803,
+ "asm": 7804,
+ "ims": 7805,
+ "▁records": 7806,
+ "▁corpor": 7807,
+ "▁intellig": 7808,
+ "▁Prem": 7809,
+ "▁driving": 7810,
+ "▁marriage": 7811,
+ "▁Thank": 7812,
+ "▁willing": 7813,
+ "MC": 7814,
+ "Fields": 7815,
+ "Items": 7816,
+ "▁micro": 7817,
+ "▁lift": 7818,
+ "irection": 7819,
+ "Account": 7820,
+ "▁architect": 7821,
+ "track": 7822,
+ "▁prin": 7823,
+ "PA": 7824,
+ "▁runs": 7825,
+ "▁Texas": 7826,
+ "isher": 7827,
+ "ensure": 7828,
+ "▁Both": 7829,
+ "ком": 7830,
+ "▁Color": 7831,
+ "Register": 7832,
+ "▁Joe": 7833,
+ "geq": 7834,
+ "lets": 7835,
+ "ading": 7836,
+ "▁army": 7837,
+ "▁Bank": 7838,
+ "otic": 7839,
+ "Product": 7840,
+ "import": 7841,
+ "▁Wed": 7842,
+ "▁cry": 7843,
+ "grade": 7844,
+ "dig": 7845,
+ "gal": 7846,
+ "кла": 7847,
+ "ested": 7848,
+ "ões": 7849,
+ "gers": 7850,
+ "ologie": 7851,
+ "том": 7852,
+ "razy": 7853,
+ "▁dinner": 7854,
+ "QU": 7855,
+ "▁fingers": 7856,
+ "ULE": 7857,
+ "claim": 7858,
+ "▁advantage": 7859,
+ "▁variable": 7860,
+ "▁medic": 7861,
+ "▁male": 7862,
+ "▁circum": 7863,
+ "▁мі": 7864,
+ "▁internet": 7865,
+ "WN": 7866,
+ "▁lab": 7867,
+ "azine": 7868,
+ "чно": 7869,
+ "▁loop": 7870,
+ "▁pred": 7871,
+ "▁consequ": 7872,
+ "▁balance": 7873,
+ "fortun": 7874,
+ "▁gift": 7875,
+ "▁drug": 7876,
+ "▁cash": 7877,
+ "ских": 7878,
+ "rg": 7879,
+ "istribut": 7880,
+ "▁highest": 7881,
+ "ême": 7882,
+ "emph": 7883,
+ "emon": 7884,
+ "▁performed": 7885,
+ "cut": 7886,
+ "▁closer": 7887,
+ "▁becoming": 7888,
+ "▁\"\",": 7889,
+ "star": 7890,
+ "pub": 7891,
+ "▁prepar": 7892,
+ "▁vote": 7893,
+ "ilde": 7894,
+ "▁impress": 7895,
+ "▁employees": 7896,
+ "▁einen": 7897,
+ "▁smooth": 7898,
+ "▁snow": 7899,
+ "▁purs": 7900,
+ "▁voc": 7901,
+ "▁Microsoft": 7902,
+ "PU": 7903,
+ "▁income": 7904,
+ "inos": 7905,
+ "▁operator": 7906,
+ "▁equival": 7907,
+ "▁password": 7908,
+ "ción": 7909,
+ "success": 7910,
+ "▁emp": 7911,
+ "HOUT": 7912,
+ "▁ca": 7913,
+ "flag": 7914,
+ "illy": 7915,
+ "crete": 7916,
+ "frak": 7917,
+ "▁hidden": 7918,
+ "▁\"%": 7919,
+ "ERN": 7920,
+ "рова": 7921,
+ "▁UN": 7922,
+ "roke": 7923,
+ "miss": 7924,
+ "▁split": 7925,
+ "Reference": 7926,
+ ")$,": 7927,
+ "eper": 7928,
+ "▁NO": 7929,
+ "▁square": 7930,
+ "sur": 7931,
+ "чен": 7932,
+ "ester": 7933,
+ "нь": 7934,
+ "}\"": 7935,
+ "rawn": 7936,
+ "rule": 7937,
+ "▁audience": 7938,
+ "este": 7939,
+ "ems": 7940,
+ "ICENSE": 7941,
+ "▁Ill": 7942,
+ "USE": 7943,
+ "▁bon": 7944,
+ "bur": 7945,
+ "▁sick": 7946,
+ "▁horse": 7947,
+ "▁Educ": 7948,
+ "▁benefit": 7949,
+ "▁cro": 7950,
+ "Application": 7951,
+ "▁corre": 7952,
+ "▁guarante": 7953,
+ "DATA": 7954,
+ "▁explained": 7955,
+ "TX": 7956,
+ "▁ont": 7957,
+ "▁Flor": 7958,
+ "▁reports": 7959,
+ "▁Real": 7960,
+ "uded": 7961,
+ "lean": 7962,
+ "▁citiz": 7963,
+ "▁decide": 7964,
+ "WS": 7965,
+ "▁domain": 7966,
+ "▁reflect": 7967,
+ "▁minimum": 7968,
+ "▁legs": 7969,
+ "▁smiled": 7970,
+ "fi": 7971,
+ "▁pure": 7972,
+ "▁Custom": 7973,
+ "▁essential": 7974,
+ "▁observed": 7975,
+ "Bytes": 7976,
+ "▁ctx": 7977,
+ "▁rates": 7978,
+ "mbre": 7979,
+ "▁worry": 7980,
+ ")^": 7981,
+ "▁Research": 7982,
+ "Root": 7983,
+ "Windows": 7984,
+ "ulture": 7985,
+ "▁relative": 7986,
+ "▁seu": 7987,
+ "▁nie": 7988,
+ "▁shook": 7989,
+ "iously": 7990,
+ "▁advert": 7991,
+ "See": 7992,
+ "▁Central": 7993,
+ "▁batter": 7994,
+ "▁signed": 7995,
+ "TS": 7996,
+ "oni": 7997,
+ "▁prepared": 7998,
+ "gate": 7999,
+ "▁Care": 8000,
+ "care": 8001,
+ "▁supply": 8002,
+ "Exp": 8003,
+ "bolds": 8004,
+ "▁trail": 8005,
+ "▁fish": 8006,
+ "▁units": 8007,
+ "venue": 8008,
+ "хи": 8009,
+ "▁Wood": 8010,
+ "▁category": 8011,
+ "▁ble": 8012,
+ "▁override": 8013,
+ "foo": 8014,
+ "▁influence": 8015,
+ "enth": 8016,
+ "rij": 8017,
+ "▁adapt": 8018,
+ "icians": 8019,
+ "deleted": 8020,
+ "▁vision": 8021,
+ "ctrl": 8022,
+ "Lambda": 8023,
+ "tp": 8024,
+ "mond": 8025,
+ "aturday": 8026,
+ "normal": 8027,
+ "▁thousand": 8028,
+ "▁Profess": 8029,
+ "▁disease": 8030,
+ "clip": 8031,
+ "▁гра": 8032,
+ "boldsymbol": 8033,
+ "OB": 8034,
+ "▁challenge": 8035,
+ "▁motion": 8036,
+ "▁whis": 8037,
+ "▁leaders": 8038,
+ "▁colon": 8039,
+ "▁suit": 8040,
+ "mid": 8041,
+ "ampion": 8042,
+ "ág": 8043,
+ "▁views": 8044,
+ "▁appears": 8045,
+ "ancel": 8046,
+ "▁zwe": 8047,
+ "IST": 8048,
+ "▁leaves": 8049,
+ "▁enh": 8050,
+ "Active": 8051,
+ "▁dit": 8052,
+ "ificate": 8053,
+ "matrix": 8054,
+ "Expression": 8055,
+ "Reader": 8056,
+ "▁mental": 8057,
+ "embre": 8058,
+ "▁decor": 8059,
+ "arts": 8060,
+ "▁vent": 8061,
+ "nel": 8062,
+ "lines": 8063,
+ "upid": 8064,
+ "erved": 8065,
+ "▁boys": 8066,
+ "аль": 8067,
+ "MOD": 8068,
+ "isl": 8069,
+ "▁[[": 8070,
+ "phy": 8071,
+ "▁..": 8072,
+ "▁agent": 8073,
+ "▁Services": 8074,
+ "▁iron": 8075,
+ "▁components": 8076,
+ "▁fre": 8077,
+ "ictionary": 8078,
+ "▁tests": 8079,
+ ".~\\": 8080,
+ "obs": 8081,
+ "▁Ми": 8082,
+ "▁обла": 8083,
+ "▁assess": 8084,
+ "▁Friday": 8085,
+ "▁weather": 8086,
+ "kg": 8087,
+ "стра": 8088,
+ ".}": 8089,
+ "endant": 8090,
+ "anna": 8091,
+ "▁Japanese": 8092,
+ "cmp": 8093,
+ "▁Army": 8094,
+ "onym": 8095,
+ "▁relax": 8096,
+ "dates": 8097,
+ "▁Russian": 8098,
+ "▁excellent": 8099,
+ "'))": 8100,
+ "ILITY": 8101,
+ "▁showing": 8102,
+ "▁Daniel": 8103,
+ "мя": 8104,
+ "▁Main": 8105,
+ "Phi": 8106,
+ "▁Rock": 8107,
+ "▁grew": 8108,
+ "▁yield": 8109,
+ "ière": 8110,
+ "seg": 8111,
+ "}}$": 8112,
+ "▁strict": 8113,
+ "▁vehicle": 8114,
+ "UD": 8115,
+ "AF": 8116,
+ "Sw": 8117,
+ "▁chest": 8118,
+ "▁officer": 8119,
+ "▁ear": 8120,
+ "HER": 8121,
+ "noon": 8122,
+ "▁journey": 8123,
+ "NT": 8124,
+ "▁divers": 8125,
+ "▁Finally": 8126,
+ "Found": 8127,
+ "▁AS": 8128,
+ "rik": 8129,
+ "▁constr": 8130,
+ "▁sust": 8131,
+ "account": 8132,
+ "▁walls": 8133,
+ "▁entirely": 8134,
+ "Iter": 8135,
+ "cha": 8136,
+ "ishes": 8137,
+ "IVE": 8138,
+ "▁prime": 8139,
+ "▁…": 8140,
+ "xe": 8141,
+ "uten": 8142,
+ "arse": 8143,
+ "▁Pa": 8144,
+ "pute": 8145,
+ "äl": 8146,
+ "▁protection": 8147,
+ "▁keys": 8148,
+ "May": 8149,
+ "Byte": 8150,
+ "Const": 8151,
+ "BL": 8152,
+ "▁пе": 8153,
+ "▁spl": 8154,
+ "▁clothes": 8155,
+ "ashed": 8156,
+ "Mark": 8157,
+ "ème": 8158,
+ "▁fait": 8159,
+ "▁introduced": 8160,
+ "unlock": 8161,
+ "▁Instead": 8162,
+ "ansion": 8163,
+ "region": 8164,
+ "▁Americans": 8165,
+ "▁indeed": 8166,
+ "widget": 8167,
+ "▁realize": 8168,
+ "▁fro": 8169,
+ "BIT": 8170,
+ "▁React": 8171,
+ "READ": 8172,
+ "asket": 8173,
+ "never": 8174,
+ "▁poll": 8175,
+ "icol": 8176,
+ "▁prev": 8177,
+ "▁hyp": 8178,
+ "▁Fur": 8179,
+ "cloud": 8180,
+ "▁Lee": 8181,
+ "pling": 8182,
+ "▁Child": 8183,
+ "▁ideal": 8184,
+ "Selector": 8185,
+ "STATUS": 8186,
+ "ucture": 8187,
+ "▁wine": 8188,
+ "▁possibly": 8189,
+ "▁putting": 8190,
+ "▁riv": 8191,
+ "▁wearing": 8192,
+ "▁Source": 8193,
+ "▁Cas": 8194,
+ "Changed": 8195,
+ "▁thanks": 8196,
+ "TIME": 8197,
+ "▁sport": 8198,
+ "▁Award": 8199,
+ "▁glad": 8200,
+ "▁Pass": 8201,
+ "▁Pos": 8202,
+ "sche": 8203,
+ "▁CD": 8204,
+ "▁afford": 8205,
+ "▁Women": 8206,
+ "▁District": 8207,
+ "▁identity": 8208,
+ "▁parties": 8209,
+ ":%": 8210,
+ "▁drag": 8211,
+ "▁mai": 8212,
+ "!(": 8213,
+ "langle": 8214,
+ "▁knowing": 8215,
+ "Project": 8216,
+ "▁regarding": 8217,
+ "▁Joseph": 8218,
+ "ге": 8219,
+ "▁Dar": 8220,
+ "▁Hor": 8221,
+ "▁animals": 8222,
+ "▁extension": 8223,
+ "ская": 8224,
+ "▁Han": 8225,
+ "btn": 8226,
+ "aciones": 8227,
+ "▁familiar": 8228,
+ "holder": 8229,
+ ":\r": 8230,
+ "stood": 8231,
+ "▁liked": 8232,
+ "CODE": 8233,
+ "▁enable": 8234,
+ "▁ped": 8235,
+ "iti": 8236,
+ "hab": 8237,
+ "DIR": 8238,
+ "▁beat": 8239,
+ "ті": 8240,
+ "▁Minister": 8241,
+ "▁py": 8242,
+ "Pat": 8243,
+ "▁exhib": 8244,
+ "▁Build": 8245,
+ "▁Field": 8246,
+ "ician": 8247,
+ "▁collabor": 8248,
+ "▁quarter": 8249,
+ "▁False": 8250,
+ "km": 8251,
+ "▁virtual": 8252,
+ "owa": 8253,
+ "▁Jon": 8254,
+ "amin": 8255,
+ "uen": 8256,
+ "▁ин": 8257,
+ "imation": 8258,
+ "oving": 8259,
+ "▁testing": 8260,
+ "sect": 8261,
+ "ITION": 8262,
+ "!\\": 8263,
+ "apy": 8264,
+ "▁transition": 8265,
+ "ository": 8266,
+ "ODO": 8267,
+ "PD": 8268,
+ "né": 8269,
+ "▁generate": 8270,
+ "▁native": 8271,
+ "▁('": 8272,
+ "▁elle": 8273,
+ "RR": 8274,
+ "▁hun": 8275,
+ "_->": 8276,
+ "agnost": 8277,
+ "▁proposed": 8278,
+ "▁Game": 8279,
+ "▁efforts": 8280,
+ "вя": 8281,
+ "tc": 8282,
+ "ск": 8283,
+ "▁intent": 8284,
+ "▁Bre": 8285,
+ "isc": 8286,
+ "▁protest": 8287,
+ "▁holds": 8288,
+ "ometry": 8289,
+ "▁Have": 8290,
+ "▁detail": 8291,
+ "▁WITHOUT": 8292,
+ "yer": 8293,
+ "▁Kon": 8294,
+ "▁noticed": 8295,
+ "▁requirements": 8296,
+ "DEBUG": 8297,
+ "kins": 8298,
+ "▁Span": 8299,
+ "▁cars": 8300,
+ "meta": 8301,
+ "▁kil": 8302,
+ "▁Bron": 8303,
+ "▁experienced": 8304,
+ "▁remind": 8305,
+ "ourse": 8306,
+ "▁Western": 8307,
+ "tered": 8308,
+ "▁devices": 8309,
+ "▁pictures": 8310,
+ "▁tut": 8311,
+ "\"`": 8312,
+ "▁impossible": 8313,
+ "▁rail": 8314,
+ "▁feels": 8315,
+ "icas": 8316,
+ "illing": 8317,
+ "▁accident": 8318,
+ "▁'@": 8319,
+ "________": 8320,
+ "▁notes": 8321,
+ "oman": 8322,
+ "Parser": 8323,
+ "▁discovered": 8324,
+ "▁Roman": 8325,
+ "▁budget": 8326,
+ "▁guide": 8327,
+ "king": 8328,
+ "▁incred": 8329,
+ "olar": 8330,
+ "enden": 8331,
+ "Desc": 8332,
+ "▁wave": 8333,
+ "бли": 8334,
+ "igt": 8335,
+ "▁restrict": 8336,
+ "▁Ret": 8337,
+ "▁mac": 8338,
+ "ур": 8339,
+ "BS": 8340,
+ "ís": 8341,
+ "▁generation": 8342,
+ "dem": 8343,
+ "alo": 8344,
+ "бра": 8345,
+ "▁ordered": 8346,
+ "drop": 8347,
+ "▁pp": 8348,
+ "▁Review": 8349,
+ "▁literally": 8350,
+ "▁Sir": 8351,
+ "▁Yeah": 8352,
+ "▁density": 8353,
+ "riz": 8354,
+ "inde": 8355,
+ "▁gain": 8356,
+ "▁panel": 8357,
+ "jet": 8358,
+ "▁Times": 8359,
+ "▁nella": 8360,
+ "▁previously": 8361,
+ "points": 8362,
+ "Send": 8363,
+ "▁Brown": 8364,
+ "each": 8365,
+ "▁trigger": 8366,
+ "ometimes": 8367,
+ "icos": 8368,
+ "GR": 8369,
+ "Panel": 8370,
+ "ogen": 8371,
+ "▁cm": 8372,
+ "ructions": 8373,
+ "▁kiss": 8374,
+ "▁solo": 8375,
+ "▁famous": 8376,
+ "ran": 8377,
+ "про": 8378,
+ "▁thro": 8379,
+ "Graph": 8380,
+ "imit": 8381,
+ "▁Value": 8382,
+ "▁starts": 8383,
+ "ipeline": 8384,
+ "hd": 8385,
+ "TC": 8386,
+ "▁discussion": 8387,
+ "▁truck": 8388,
+ "aka": 8389,
+ "Only": 8390,
+ "▁Equ": 8391,
+ "▁kö": 8392,
+ "▁Bes": 8393,
+ "▁critic": 8394,
+ "▁propos": 8395,
+ "▁batt": 8396,
+ "▁Section": 8397,
+ "Show": 8398,
+ "gp": 8399,
+ "STATE": 8400,
+ "POST": 8401,
+ "▁Nord": 8402,
+ "▁innov": 8403,
+ "▁crim": 8404,
+ "axis": 8405,
+ "▁Turn": 8406,
+ "conn": 8407,
+ "Runtime": 8408,
+ "▁remaining": 8409,
+ "oston": 8410,
+ "▁Э": 8411,
+ "▁windows": 8412,
+ "▁Royal": 8413,
+ "▁vide": 8414,
+ "PP": 8415,
+ "chron": 8416,
+ "▁san": 8417,
+ "▁rise": 8418,
+ "▁delle": 8419,
+ "▁Dur": 8420,
+ "▁rapid": 8421,
+ "cert": 8422,
+ "LA": 8423,
+ "edge": 8424,
+ "▁\\]": 8425,
+ "▁entered": 8426,
+ "▁laws": 8427,
+ "▁photo": 8428,
+ "▁applications": 8429,
+ "▁Berlin": 8430,
+ "▁arrest": 8431,
+ "▁federal": 8432,
+ "▁Russia": 8433,
+ "▁usual": 8434,
+ "▁raw": 8435,
+ "▁più": 8436,
+ "être": 8437,
+ "JSON": 8438,
+ "SION": 8439,
+ "xture": 8440,
+ "istent": 8441,
+ "▁Power": 8442,
+ "Bit": 8443,
+ "▁capacity": 8444,
+ "▁cards": 8445,
+ "UID": 8446,
+ "iments": 8447,
+ "▁dar": 8448,
+ "▁Chicago": 8449,
+ "▁comfortable": 8450,
+ "tip": 8451,
+ "bas": 8452,
+ "▁mu": 8453,
+ "▁enemy": 8454,
+ "yan": 8455,
+ "▁фи": 8456,
+ "▁updated": 8457,
+ "ango": 8458,
+ "Ev": 8459,
+ "Effect": 8460,
+ "osing": 8461,
+ "rence": 8462,
+ "▁Congress": 8463,
+ "▁defe": 8464,
+ "▁ip": 8465,
+ "▁tout": 8466,
+ "▁freedom": 8467,
+ "▁ao": 8468,
+ "▁Therefore": 8469,
+ "Edit": 8470,
+ "▁Virgin": 8471,
+ "REE": 8472,
+ "argo": 8473,
+ "▁Dam": 8474,
+ "▁traffic": 8475,
+ "ños": 8476,
+ "▁alle": 8477,
+ "▁depth": 8478,
+ "Now": 8479,
+ "▁sides": 8480,
+ "▁годи": 8481,
+ "Descriptor": 8482,
+ "▁artikel": 8483,
+ "▁narrow": 8484,
+ "___": 8485,
+ "kw": 8486,
+ "uto": 8487,
+ "▁Facebook": 8488,
+ "tegr": 8489,
+ "boolean": 8490,
+ "nik": 8491,
+ "bd": 8492,
+ "Track": 8493,
+ "▁gran": 8494,
+ "reshold": 8495,
+ "вет": 8496,
+ "wrap": 8497,
+ "▁noise": 8498,
+ "igu": 8499,
+ "▁Bon": 8500,
+ "▁wy": 8501,
+ "linux": 8502,
+ "cks": 8503,
+ "▁fans": 8504,
+ "▁mach": 8505,
+ "▁prices": 8506,
+ "év": 8507,
+ "outs": 8508,
+ "standing": 8509,
+ "▁categ": 8510,
+ ";\\": 8511,
+ "▁decre": 8512,
+ "▁Saturday": 8513,
+ "▁menu": 8514,
+ "▁Nov": 8515,
+ "▁Yet": 8516,
+ "▁так": 8517,
+ "liche": 8518,
+ "▁Academ": 8519,
+ "▁communication": 8520,
+ "using": 8521,
+ "▁Society": 8522,
+ "▁nuc": 8523,
+ "pective": 8524,
+ "orial": 8525,
+ "▁afraid": 8526,
+ "▁animal": 8527,
+ "▁turning": 8528,
+ "dst": 8529,
+ "mathfrak": 8530,
+ "lers": 8531,
+ "▁lots": 8532,
+ "▁á": 8533,
+ "▁Tra": 8534,
+ "np": 8535,
+ "▁rose": 8536,
+ "▁GL": 8537,
+ "▁helping": 8538,
+ "▁winter": 8539,
+ "▁ком": 8540,
+ "Mock": 8541,
+ "▁investment": 8542,
+ "Use": 8543,
+ "▁Canad": 8544,
+ "нд": 8545,
+ "Copy": 8546,
+ "▁fly": 8547,
+ "SER": 8548,
+ "▁Far": 8549,
+ "▁Ros": 8550,
+ "amil": 8551,
+ "▁fighting": 8552,
+ "▁religious": 8553,
+ "super": 8554,
+ "screen": 8555,
+ "▁furn": 8556,
+ "▁surprised": 8557,
+ "▁replied": 8558,
+ "Activity": 8559,
+ "▁Down": 8560,
+ "▁insert": 8561,
+ "▁Olymp": 8562,
+ "▁pointed": 8563,
+ "▁Card": 8564,
+ "driver": 8565,
+ "▁Da": 8566,
+ "!--": 8567,
+ "roud": 8568,
+ "undo": 8569,
+ "▁messages": 8570,
+ "▁Point": 8571,
+ "VM": 8572,
+ "▁plane": 8573,
+ "xc": 8574,
+ "▁television": 8575,
+ "ён": 8576,
+ "▁thousands": 8577,
+ "▁cris": 8578,
+ "▁delay": 8579,
+ "▁Next": 8580,
+ "▁nombre": 8581,
+ "▁tu": 8582,
+ "▁skip": 8583,
+ "road": 8584,
+ "istration": 8585,
+ "▁tur": 8586,
+ "▁Develop": 8587,
+ "▁Па": 8588,
+ "▁дру": 8589,
+ "▁wonderful": 8590,
+ ">&": 8591,
+ "▁Liber": 8592,
+ "▁scope": 8593,
+ "▁manage": 8594,
+ "▁dass": 8595,
+ "▁recall": 8596,
+ "PM": 8597,
+ "▁relevant": 8598,
+ "▁Earth": 8599,
+ "▁как": 8600,
+ "▁apr": 8601,
+ "▁ASS": 8602,
+ "ién": 8603,
+ "▁SH": 8604,
+ "oom": 8605,
+ "itet": 8606,
+ "none": 8607,
+ "asi": 8608,
+ "▁motor": 8609,
+ "▁Show": 8610,
+ "nb": 8611,
+ "▁factors": 8612,
+ "▁forest": 8613,
+ "▁вре": 8614,
+ "thm": 8615,
+ "▁municip": 8616,
+ "▁turns": 8617,
+ "▁Division": 8618,
+ "EC": 8619,
+ "▁disappe": 8620,
+ "structor": 8621,
+ "▁somewhere": 8622,
+ "▁African": 8623,
+ "▁Institute": 8624,
+ "Grid": 8625,
+ "▁teacher": 8626,
+ "uries": 8627,
+ "▁respectively": 8628,
+ "▁SD": 8629,
+ "▁alive": 8630,
+ "▁pou": 8631,
+ "▁Water": 8632,
+ "фе": 8633,
+ "▁changing": 8634,
+ "▁afternoon": 8635,
+ "▁orders": 8636,
+ "Ret": 8637,
+ "Pointer": 8638,
+ "▁sav": 8639,
+ "erg": 8640,
+ "oked": 8641,
+ "essions": 8642,
+ "▁Fire": 8643,
+ "aret": 8644,
+ "imm": 8645,
+ "▁desire": 8646,
+ "▁що": 8647,
+ "▁Design": 8648,
+ "uture": 8649,
+ "▁Office": 8650,
+ "▁cmd": 8651,
+ "▁eating": 8652,
+ "Network": 8653,
+ "▁rough": 8654,
+ "operator": 8655,
+ "IGN": 8656,
+ "▁sports": 8657,
+ "▁weren": 8658,
+ "▁noted": 8659,
+ "▁twice": 8660,
+ "III": 8661,
+ "▁anx": 8662,
+ "▁elim": 8663,
+ "▁ав": 8664,
+ "▁io": 8665,
+ "▁speech": 8666,
+ "▁condu": 8667,
+ "elles": 8668,
+ "idade": 8669,
+ "▁advance": 8670,
+ "RI": 8671,
+ "oca": 8672,
+ "/\\": 8673,
+ "apshot": 8674,
+ "▁tail": 8675,
+ "models": 8676,
+ "ogy": 8677,
+ "▁Jeff": 8678,
+ "iration": 8679,
+ "▁Kore": 8680,
+ "▁leads": 8681,
+ "bat": 8682,
+ "Adapter": 8683,
+ "category": 8684,
+ "angular": 8685,
+ "▁saved": 8686,
+ "▁uniform": 8687,
+ "▁né": 8688,
+ "▁businesses": 8689,
+ "Hist": 8690,
+ "▁ар": 8691,
+ "domain": 8692,
+ "▁Si": 8693,
+ "raise": 8694,
+ "▁warn": 8695,
+ "hetic": 8696,
+ "▁Gro": 8697,
+ ")).": 8698,
+ "}>": 8699,
+ "зе": 8700,
+ "▁Amazon": 8701,
+ "▁Organ": 8702,
+ "▁Lake": 8703,
+ "▁agreement": 8704,
+ "xa": 8705,
+ "▁perman": 8706,
+ "▁containing": 8707,
+ "▁strange": 8708,
+ "сті": 8709,
+ "▁stupid": 8710,
+ "▁speaking": 8711,
+ "▁Internet": 8712,
+ "prefix": 8713,
+ "esc": 8714,
+ "Assert": 8715,
+ "prote": 8716,
+ "▁manner": 8717,
+ "▁Sz": 8718,
+ "unte": 8719,
+ "iot": 8720,
+ "Profile": 8721,
+ "oven": 8722,
+ "▁formed": 8723,
+ "▁lit": 8724,
+ "▁economy": 8725,
+ "▁cz": 8726,
+ "wid": 8727,
+ "REQ": 8728,
+ "▁chosen": 8729,
+ "▁Produ": 8730,
+ "oster": 8731,
+ "stances": 8732,
+ "awa": 8733,
+ "▁Ren": 8734,
+ "▁confirm": 8735,
+ "▁Бо": 8736,
+ "▁billion": 8737,
+ "▁déc": 8738,
+ "ých": 8739,
+ "▁illustr": 8740,
+ "TIES": 8741,
+ "▁Pub": 8742,
+ "▁ban": 8743,
+ "aded": 8744,
+ "ahn": 8745,
+ "▁Cath": 8746,
+ "nonumber": 8747,
+ "▁worst": 8748,
+ "▁Ме": 8749,
+ "▁suggested": 8750,
+ "stats": 8751,
+ "▁cant": 8752,
+ "▁align": 8753,
+ "kappa": 8754,
+ "▁hen": 8755,
+ "▁initi": 8756,
+ "'])": 8757,
+ "BI": 8758,
+ "▁garden": 8759,
+ "▁secure": 8760,
+ "▁\\[": 8761,
+ "handler": 8762,
+ "elli": 8763,
+ "ldots": 8764,
+ "secut": 8765,
+ "▁extended": 8766,
+ "}-": 8767,
+ "anie": 8768,
+ "▁Find": 8769,
+ "▁Museum": 8770,
+ "▁Conne": 8771,
+ "yy": 8772,
+ "▁passion": 8773,
+ "akers": 8774,
+ "ahr": 8775,
+ "ologies": 8776,
+ "▁equation": 8777,
+ "▁occasion": 8778,
+ "Let": 8779,
+ "']['": 8780,
+ "Print": 8781,
+ "anes": 8782,
+ "iente": 8783,
+ "▁Today": 8784,
+ "LECT": 8785,
+ "▁Af": 8786,
+ ",,": 8787,
+ "▁Та": 8788,
+ "▁```": 8789,
+ "even": 8790,
+ "sin": 8791,
+ "urer": 8792,
+ "▁°": 8793,
+ "otimes": 8794,
+ "▁IO": 8795,
+ "▁poet": 8796,
+ "()));": 8797,
+ "▁−": 8798,
+ "▁adopt": 8799,
+ "phere": 8800,
+ "#[": 8801,
+ "▁centre": 8802,
+ "oves": 8803,
+ "▁ans": 8804,
+ "dp": 8805,
+ "▁Kir": 8806,
+ "▁applicable": 8807,
+ "fp": 8808,
+ "▁visual": 8809,
+ "▁okay": 8810,
+ "oro": 8811,
+ "▁opportunities": 8812,
+ "Repository": 8813,
+ "▁ll": 8814,
+ "▁Rod": 8815,
+ "▁shel": 8816,
+ "▁launch": 8817,
+ "▁conven": 8818,
+ "▁Spe": 8819,
+ "Amer": 8820,
+ "▁cette": 8821,
+ "Cond": 8822,
+ "dep": 8823,
+ "Own": 8824,
+ "▁hook": 8825,
+ "▁dict": 8826,
+ "▁Those": 8827,
+ "▁fellow": 8828,
+ "▁philosoph": 8829,
+ "vin": 8830,
+ "ferences": 8831,
+ "hav": 8832,
+ "▁adding": 8833,
+ "iverse": 8834,
+ "game": 8835,
+ "▁Blue": 8836,
+ "▁clin": 8837,
+ "note": 8838,
+ "▁Ram": 8839,
+ "мер": 8840,
+ "covery": 8841,
+ "ña": 8842,
+ "▁би": 8843,
+ "▁fashion": 8844,
+ "▁broke": 8845,
+ "▁'\\": 8846,
+ "▁reader": 8847,
+ "ное": 8848,
+ "ности": 8849,
+ "▁payment": 8850,
+ "▁Lic": 8851,
+ "▁lips": 8852,
+ "▁academ": 8853,
+ "▁Mot": 8854,
+ "ells": 8855,
+ "CHECK": 8856,
+ "▁ру": 8857,
+ "▁MS": 8858,
+ "Editor": 8859,
+ "▁zone": 8860,
+ "iture": 8861,
+ "▁IT": 8862,
+ "runtime": 8863,
+ "▁proceed": 8864,
+ "лов": 8865,
+ "▁Maria": 8866,
+ "olver": 8867,
+ "▁Thanks": 8868,
+ "▁shouldn": 8869,
+ "▁Joh": 8870,
+ "▁Model": 8871,
+ "▁Sov": 8872,
+ "!'": 8873,
+ "Di": 8874,
+ "▁cancer": 8875,
+ "Ident": 8876,
+ "▁exchange": 8877,
+ "iller": 8878,
+ "inf": 8879,
+ "LEN": 8880,
+ "(){": 8881,
+ "aga": 8882,
+ "\"],": 8883,
+ "uh": 8884,
+ "▁Ken": 8885,
+ "▁photos": 8886,
+ "▁tiny": 8887,
+ "▁gent": 8888,
+ "ül": 8889,
+ "▁Take": 8890,
+ "idel": 8891,
+ "outing": 8892,
+ "Internal": 8893,
+ "▁cells": 8894,
+ "ним": 8895,
+ "hard": 8896,
+ "▁Town": 8897,
+ "obe": 8898,
+ "plex": 8899,
+ "тер": 8900,
+ "tons": 8901,
+ "▁concentr": 8902,
+ "mock": 8903,
+ "vc": 8904,
+ "áz": 8905,
+ "▁Championship": 8906,
+ "▁бе": 8907,
+ "??": 8908,
+ "éri": 8909,
+ "aly": 8910,
+ "▁Ц": 8911,
+ "ierte": 8912,
+ "▁totally": 8913,
+ "▁Auf": 8914,
+ "▁ourselves": 8915,
+ "▁Self": 8916,
+ "Forms": 8917,
+ "ighter": 8918,
+ "▁island": 8919,
+ "fmt": 8920,
+ "▁rc": 8921,
+ "▁tells": 8922,
+ "BB": 8923,
+ "dit": 8924,
+ "▁variables": 8925,
+ "▁intended": 8926,
+ "izont": 8927,
+ "▁plays": 8928,
+ "dam": 8929,
+ "seq": 8930,
+ "▁Sup": 8931,
+ "▁cultural": 8932,
+ "▁scream": 8933,
+ "__,": 8934,
+ "cipl": 8935,
+ "Timeout": 8936,
+ "▁ж": 8937,
+ "orte": 8938,
+ "▁replaced": 8939,
+ "EM": 8940,
+ "▁abandon": 8941,
+ "▁Special": 8942,
+ "ellen": 8943,
+ "▁Bru": 8944,
+ "irmed": 8945,
+ "Te": 8946,
+ "olt": 8947,
+ "ju": 8948,
+ "Argument": 8949,
+ "▁neut": 8950,
+ "scape": 8951,
+ "▁Ray": 8952,
+ "▁Polit": 8953,
+ "▁crowd": 8954,
+ "▁Windows": 8955,
+ "iego": 8956,
+ "▁escape": 8957,
+ "▁Apache": 8958,
+ "sync": 8959,
+ "eben": 8960,
+ "ifies": 8961,
+ "ether": 8962,
+ "Meta": 8963,
+ "▁biggest": 8964,
+ "Game": 8965,
+ "▁transaction": 8966,
+ "Env": 8967,
+ "▁Мо": 8968,
+ "▁plenty": 8969,
+ "▁mel": 8970,
+ "пре": 8971,
+ "▁motiv": 8972,
+ "▁ор": 8973,
+ "organ": 8974,
+ "▁mock": 8975,
+ "▁$_": 8976,
+ "ене": 8977,
+ "▁Number": 8978,
+ "cknow": 8979,
+ "▁Update": 8980,
+ "zero": 8981,
+ "▁surprise": 8982,
+ "cean": 8983,
+ "pdf": 8984,
+ "Global": 8985,
+ "▁attend": 8986,
+ "▁fond": 8987,
+ "▁understood": 8988,
+ "Nav": 8989,
+ "▁Mic": 8990,
+ "=$": 8991,
+ "oking": 8992,
+ "▁Stadium": 8993,
+ "Close": 8994,
+ "▁competition": 8995,
+ "▁soldiers": 8996,
+ "▁OP": 8997,
+ "agne": 8998,
+ "▁Anton": 8999,
+ "Main": 9000,
+ "ák": 9001,
+ "▁#[": 9002,
+ "▁Commit": 9003,
+ "pyx": 9004,
+ "▁east": 9005,
+ "▁Order": 9006,
+ "Float": 9007,
+ "▁accepted": 9008,
+ "▁monitor": 9009,
+ "▁pad": 9010,
+ "onic": 9011,
+ "▁pushed": 9012,
+ "▁replace": 9013,
+ "CRE": 9014,
+ "▁ride": 9015,
+ "found": 9016,
+ "=%": 9017,
+ "вой": 9018,
+ "▁matches": 9019,
+ "▁Lie": 9020,
+ "▁experiences": 9021,
+ "Pool": 9022,
+ "ups": 9023,
+ "AV": 9024,
+ "▁existence": 9025,
+ "▁thin": 9026,
+ "▁magn": 9027,
+ "COMP": 9028,
+ "home": 9029,
+ "▁ni": 9030,
+ "▁wurden": 9031,
+ "лав": 9032,
+ "▁teeth": 9033,
+ "▁Stan": 9034,
+ "appro": 9035,
+ "anny": 9036,
+ "ifts": 9037,
+ "▁unknown": 9038,
+ "▁homes": 9039,
+ "▁entity": 9040,
+ "cie": 9041,
+ "ление": 9042,
+ "iar": 9043,
+ "▁compliance": 9044,
+ "▁focused": 9045,
+ "uzz": 9046,
+ "=\\\"": 9047,
+ "components": 9048,
+ "Attr": 9049,
+ "allery": 9050,
+ "▁identify": 9051,
+ "Ok": 9052,
+ "pie": 9053,
+ "▁Still": 9054,
+ "▁offering": 9055,
+ "▁busy": 9056,
+ "ctl": 9057,
+ "itors": 9058,
+ "▁concerned": 9059,
+ "▁brown": 9060,
+ "clk": 9061,
+ "Selected": 9062,
+ "▁Block": 9063,
+ "▁egy": 9064,
+ "icing": 9065,
+ "▁URL": 9066,
+ "▁topic": 9067,
+ "▁Product": 9068,
+ "▁чи": 9069,
+ "▁trial": 9070,
+ "▁weekend": 9071,
+ "lu": 9072,
+ "▁IV": 9073,
+ "▁Egy": 9074,
+ "xC": 9075,
+ "▁nove": 9076,
+ "▁lett": 9077,
+ "enne": 9078,
+ "()).": 9079,
+ ".**": 9080,
+ "▁promise": 9081,
+ "election": 9082,
+ "Auth": 9083,
+ "rv": 9084,
+ "ril": 9085,
+ "▁conduct": 9086,
+ "▁maintain": 9087,
+ "▁boat": 9088,
+ "▁opposite": 9089,
+ "spin": 9090,
+ "webpack": 9091,
+ "anta": 9092,
+ "▁orient": 9093,
+ "▁suc": 9094,
+ "▁exercise": 9095,
+ "▁efficient": 9096,
+ "▁tradition": 9097,
+ "▁zw": 9098,
+ "▁Sud": 9099,
+ "going": 9100,
+ "▁Pier": 9101,
+ "inv": 9102,
+ "ipes": 9103,
+ "ensuremath": 9104,
+ "▁conver": 9105,
+ "creen": 9106,
+ "▁terror": 9107,
+ "▁Dou": 9108,
+ "▁invalid": 9109,
+ "ceived": 9110,
+ "▁Arab": 9111,
+ "▁wire": 9112,
+ "application": 9113,
+ "shift": 9114,
+ "Generic": 9115,
+ "▁Plan": 9116,
+ "▁Wall": 9117,
+ "▁directory": 9118,
+ "▁egg": 9119,
+ "▁wealth": 9120,
+ "random": 9121,
+ "attribute": 9122,
+ "▁hide": 9123,
+ "Serial": 9124,
+ "cam": 9125,
+ "▁ital": 9126,
+ "▁Line": 9127,
+ "▁CHECK": 9128,
+ "ployment": 9129,
+ "▁massive": 9130,
+ "▁extract": 9131,
+ "chain": 9132,
+ "Rest": 9133,
+ "▁Las": 9134,
+ "▁bear": 9135,
+ "▁links": 9136,
+ "▁newsp": 9137,
+ "▁FC": 9138,
+ "Card": 9139,
+ "aks": 9140,
+ "▁visible": 9141,
+ "▁Marc": 9142,
+ "▁Boston": 9143,
+ "▁reserved": 9144,
+ "▁roof": 9145,
+ "licenses": 9146,
+ "dc": 9147,
+ "▁Information": 9148,
+ "▁witness": 9149,
+ "Sk": 9150,
+ "*),": 9151,
+ "Scope": 9152,
+ "'];": 9153,
+ "▁Mir": 9154,
+ "uding": 9155,
+ "▁trend": 9156,
+ "rep": 9157,
+ "▁musical": 9158,
+ "▁neither": 9159,
+ "▁Creat": 9160,
+ "▁positions": 9161,
+ "LC": 9162,
+ "ridge": 9163,
+ "▁officers": 9164,
+ "▁violence": 9165,
+ "▁Tem": 9166,
+ "▁Sus": 9167,
+ "▁Way": 9168,
+ "After": 9169,
+ "acket": 9170,
+ "▁Sou": 9171,
+ "acer": 9172,
+ "||": 9173,
+ "▁remark": 9174,
+ "water": 9175,
+ "ně": 9176,
+ "▁Са": 9177,
+ "▁sed": 9178,
+ "Each": 9179,
+ "▁photograph": 9180,
+ "▁letters": 9181,
+ "▁invent": 9182,
+ "▁Mas": 9183,
+ "▁songs": 9184,
+ "ól": 9185,
+ "kind": 9186,
+ "▁Non": 9187,
+ "▁dust": 9188,
+ "**:": 9189,
+ "nabla": 9190,
+ ".\",": 9191,
+ "Lock": 9192,
+ "▁До": 9193,
+ "▁cluster": 9194,
+ "loss": 9195,
+ "▁ASSERT": 9196,
+ "fall": 9197,
+ "▁reject": 9198,
+ "▁Spring": 9199,
+ "▁wedding": 9200,
+ "▁grav": 9201,
+ "ression": 9202,
+ "limit": 9203,
+ "RES": 9204,
+ "]}": 9205,
+ "▁listed": 9206,
+ "▁Tele": 9207,
+ "hline": 9208,
+ "▁chief": 9209,
+ "MEM": 9210,
+ "дар": 9211,
+ "▁expensive": 9212,
+ "trace": 9213,
+ "▁Rog": 9214,
+ "▁Coll": 9215,
+ "▁Author": 9216,
+ "▁Board": 9217,
+ "▁Capt": 9218,
+ "TEXT": 9219,
+ "▁recon": 9220,
+ "esta": 9221,
+ "▁properly": 9222,
+ "▁&\\": 9223,
+ "leton": 9224,
+ "iker": 9225,
+ "Gu": 9226,
+ "▁Kom": 9227,
+ "oco": 9228,
+ "▁anymore": 9229,
+ "▁taste": 9230,
+ "▁Santa": 9231,
+ "gex": 9232,
+ "▁Secret": 9233,
+ "▁talent": 9234,
+ "▁moments": 9235,
+ "▁Ba": 9236,
+ "▁extr": 9237,
+ "▁Commission": 9238,
+ "▁modify": 9239,
+ "▁Figure": 9240,
+ "▁domin": 9241,
+ "▁plot": 9242,
+ "enger": 9243,
+ "utch": 9244,
+ "▁cities": 9245,
+ "▁nut": 9246,
+ "profile": 9247,
+ "▁Stat": 9248,
+ "▁nodes": 9249,
+ "▁ns": 9250,
+ "essages": 9251,
+ "impl": 9252,
+ "icker": 9253,
+ "▁examples": 9254,
+ "abeth": 9255,
+ "▁stated": 9256,
+ "fire": 9257,
+ "bul": 9258,
+ "▁dangerous": 9259,
+ "▁Pay": 9260,
+ "▁Gre": 9261,
+ "▁Monday": 9262,
+ "esome": 9263,
+ "igan": 9264,
+ "rund": 9265,
+ "prise": 9266,
+ "fail": 9267,
+ "▁Never": 9268,
+ "Av": 9269,
+ "▁linear": 9270,
+ "▁ul": 9271,
+ "WAR": 9272,
+ "рен": 9273,
+ "▁AT": 9274,
+ "▁dop": 9275,
+ "▁nou": 9276,
+ "Dest": 9277,
+ "▁claims": 9278,
+ "enda": 9279,
+ "▁crazy": 9280,
+ "gel": 9281,
+ "oggle": 9282,
+ "▁representation": 9283,
+ "inen": 9284,
+ "▁alternative": 9285,
+ "DM": 9286,
+ "ABILITY": 9287,
+ "faces": 9288,
+ "▁doors": 9289,
+ "ativ": 9290,
+ "Look": 9291,
+ "▁JSON": 9292,
+ "▁appearance": 9293,
+ "бря": 9294,
+ "SQL": 9295,
+ "▁silence": 9296,
+ "udo": 9297,
+ "▁Director": 9298,
+ "Statement": 9299,
+ "selected": 9300,
+ "high": 9301,
+ "prime": 9302,
+ "▁ignore": 9303,
+ "▁colors": 9304,
+ "ushing": 9305,
+ "▁virt": 9306,
+ "manager": 9307,
+ "▁remote": 9308,
+ "ło": 9309,
+ "small": 9310,
+ "▁crime": 9311,
+ "rb": 9312,
+ "▁creation": 9313,
+ "▁flight": 9314,
+ "▁Sign": 9315,
+ "ILE": 9316,
+ "▁DO": 9317,
+ "comment": 9318,
+ "▁Cost": 9319,
+ ".__": 9320,
+ "▁Cop": 9321,
+ "▁vom": 9322,
+ "▁Science": 9323,
+ "ления": 9324,
+ "oop": 9325,
+ "interface": 9326,
+ "▁WARRANTIES": 9327,
+ "▁Page": 9328,
+ "******": 9329,
+ "ском": 9330,
+ "TRUE": 9331,
+ "▁repeated": 9332,
+ "▁его": 9333,
+ "шо": 9334,
+ "▁roz": 9335,
+ "Pe": 9336,
+ "▁ISBN": 9337,
+ "irts": 9338,
+ "poses": 9339,
+ "})$": 9340,
+ "▁І": 9341,
+ "children": 9342,
+ "bles": 9343,
+ "ECT": 9344,
+ "▁iz": 9345,
+ "▁builder": 9346,
+ "▁Media": 9347,
+ "iat": 9348,
+ "▁contrast": 9349,
+ "”,": 9350,
+ "▁Link": 9351,
+ "▁Education": 9352,
+ "▁joint": 9353,
+ "▁external": 9354,
+ "▁роз": 9355,
+ "▁bits": 9356,
+ "FORM": 9357,
+ "erman": 9358,
+ "wp": 9359,
+ "▁Mike": 9360,
+ "▁Master": 9361,
+ "▁senior": 9362,
+ "▁Nav": 9363,
+ "▁recorded": 9364,
+ "eling": 9365,
+ "esh": 9366,
+ "fx": 9367,
+ "кан": 9368,
+ "▁tall": 9369,
+ "▁Johnson": 9370,
+ "▁sono": 9371,
+ "▁anche": 9372,
+ "icken": 9373,
+ "loop": 9374,
+ "iciency": 9375,
+ "emporary": 9376,
+ "▁Does": 9377,
+ "▁relation": 9378,
+ "мы": 9379,
+ "was": 9380,
+ "low": 9381,
+ "ichte": 9382,
+ "▁Jones": 9383,
+ "▁bedroom": 9384,
+ "DIS": 9385,
+ "▁magnet": 9386,
+ "▁Engine": 9387,
+ "▁feelings": 9388,
+ "GC": 9389,
+ "▁torn": 9390,
+ "▁relationships": 9391,
+ "▁Ре": 9392,
+ "▁proud": 9393,
+ "▁twe": 9394,
+ "oval": 9395,
+ "▁waste": 9396,
+ "▁reduced": 9397,
+ "ilton": 9398,
+ "BP": 9399,
+ "▁forgot": 9400,
+ "▁bodies": 9401,
+ "▁Haw": 9402,
+ "lag": 9403,
+ "▁www": 9404,
+ "door": 9405,
+ "▁sufficient": 9406,
+ "▁dollars": 9407,
+ "Len": 9408,
+ "▁talked": 9409,
+ "▁bond": 9410,
+ "▁Bor": 9411,
+ "}}{": 9412,
+ "rod": 9413,
+ "Password": 9414,
+ "quare": 9415,
+ "▁lights": 9416,
+ "eren": 9417,
+ "▁thirty": 9418,
+ "NC": 9419,
+ "▁TODO": 9420,
+ "▁respond": 9421,
+ "ких": 9422,
+ "direct": 9423,
+ "ação": 9424,
+ "▁heav": 9425,
+ "Media": 9426,
+ "exit": 9427,
+ "License": 9428,
+ "`.": 9429,
+ "▁mixed": 9430,
+ "▁desk": 9431,
+ "▁teaching": 9432,
+ "▁maj": 9433,
+ "▁nerv": 9434,
+ "inations": 9435,
+ "typeof": 9436,
+ "▁coast": 9437,
+ "▁же": 9438,
+ "▁beside": 9439,
+ "ummy": 9440,
+ "Doc": 9441,
+ "▁schedule": 9442,
+ "▁recover": 9443,
+ "▁Further": 9444,
+ "▁steel": 9445,
+ "boot": 9446,
+ "▁Perhaps": 9447,
+ "▁съ": 9448,
+ "▁Os": 9449,
+ "rick": 9450,
+ "▁Ви": 9451,
+ "Support": 9452,
+ "▁(_": 9453,
+ "nil": 9454,
+ "pis": 9455,
+ "xpected": 9456,
+ "▁processing": 9457,
+ "Build": 9458,
+ "arian": 9459,
+ "▁icon": 9460,
+ "▁CA": 9461,
+ "wick": 9462,
+ "=(": 9463,
+ "▁algorithm": 9464,
+ "▁Young": 9465,
+ "▁Management": 9466,
+ "▁ancient": 9467,
+ "ность": 9468,
+ "oti": 9469,
+ "▁combination": 9470,
+ "world": 9471,
+ "nn": 9472,
+ "▁dram": 9473,
+ "enabled": 9474,
+ "Ac": 9475,
+ "CCESS": 9476,
+ "aration": 9477,
+ "▁blocks": 9478,
+ "▁Angeles": 9479,
+ "▁Qual": 9480,
+ "▁succeed": 9481,
+ "network": 9482,
+ "▁oblig": 9483,
+ "springframework": 9484,
+ "▁Tre": 9485,
+ "okes": 9486,
+ "mun": 9487,
+ "▁Network": 9488,
+ "Del": 9489,
+ "▁estate": 9490,
+ "▁liqu": 9491,
+ "▁pob": 9492,
+ "▁dad": 9493,
+ "▁distinct": 9494,
+ "▁Tit": 9495,
+ "▁Lear": 9496,
+ "ferred": 9497,
+ "android": 9498,
+ "▁subsequ": 9499,
+ "▁Florida": 9500,
+ "subset": 9501,
+ "▁whisper": 9502,
+ "Vol": 9503,
+ "ulous": 9504,
+ "▁crew": 9505,
+ "▁lug": 9506,
+ "pid": 9507,
+ "ocity": 9508,
+ "skb": 9509,
+ "▁tea": 9510,
+ "ун": 9511,
+ "▁honor": 9512,
+ "▁Ins": 9513,
+ "▁gew": 9514,
+ "Details": 9515,
+ "eneath": 9516,
+ "atar": 9517,
+ "▁_{": 9518,
+ "amen": 9519,
+ "▁setup": 9520,
+ "Transaction": 9521,
+ "▁blank": 9522,
+ "Failed": 9523,
+ "job": 9524,
+ "▁pret": 9525,
+ "ße": 9526,
+ "loor": 9527,
+ "ří": 9528,
+ "ncia": 9529,
+ "▁anywhere": 9530,
+ "▁Light": 9531,
+ "▁Ak": 9532,
+ "BD": 9533,
+ "▁excited": 9534,
+ "agers": 9535,
+ "▁warning": 9536,
+ "▁processes": 9537,
+ "hu": 9538,
+ "▁youth": 9539,
+ "▁dogs": 9540,
+ "▁oct": 9541,
+ "▁nine": 9542,
+ "Writer": 9543,
+ "grid": 9544,
+ "▁importance": 9545,
+ "estic": 9546,
+ "▁carefully": 9547,
+ "master": 9548,
+ "▁decisions": 9549,
+ "▁pin": 9550,
+ "▁crack": 9551,
+ "TEST": 9552,
+ "▁Local": 9553,
+ "▁Right": 9554,
+ "▁vast": 9555,
+ "▁faster": 9556,
+ "▁institut": 9557,
+ "▁annual": 9558,
+ "LAN": 9559,
+ "▁episode": 9560,
+ "▁XV": 9561,
+ "▁delivery": 9562,
+ "tl": 9563,
+ "FP": 9564,
+ "circ": 9565,
+ "▁typically": 9566,
+ "igo": 9567,
+ "▁intel": 9568,
+ "nat": 9569,
+ "xb": 9570,
+ "стро": 9571,
+ ")-": 9572,
+ "▁Bal": 9573,
+ "▁Jos": 9574,
+ "▁gonna": 9575,
+ "▁Rest": 9576,
+ "jor": 9577,
+ "onia": 9578,
+ "orship": 9579,
+ "overy": 9580,
+ "LINE": 9581,
+ "]:": 9582,
+ "Queue": 9583,
+ "▁compare": 9584,
+ "▁apartment": 9585,
+ "▁rul": 9586,
+ "Dr": 9587,
+ "gency": 9588,
+ "▁obviously": 9589,
+ "zie": 9590,
+ "ycl": 9591,
+ "fortunately": 9592,
+ "▁stepped": 9593,
+ "▁Seg": 9594,
+ "▁Which": 9595,
+ "▁PC": 9596,
+ "▁ast": 9597,
+ "endor": 9598,
+ "▁permission": 9599,
+ "COL": 9600,
+ "▁TEST": 9601,
+ "Pay": 9602,
+ "ères": 9603,
+ "▁studied": 9604,
+ "▁accompl": 9605,
+ "role": 9606,
+ "Where": 9607,
+ "protobuf": 9608,
+ "metadata": 9609,
+ "Job": 9610,
+ "▁Four": 9611,
+ "plements": 9612,
+ "disable": 9613,
+ "▁loud": 9614,
+ "▁happening": 9615,
+ "▁Using": 9616,
+ "rog": 9617,
+ "▁depends": 9618,
+ "ím": 9619,
+ "'\\": 9620,
+ "▁taught": 9621,
+ "shared": 9622,
+ "▁attributes": 9623,
+ "▁Action": 9624,
+ "▁dess": 9625,
+ "▁houses": 9626,
+ "▁reset": 9627,
+ "▁bien": 9628,
+ "▁explicit": 9629,
+ "LOW": 9630,
+ "->_": 9631,
+ "▁PM": 9632,
+ "Category": 9633,
+ "oice": 9634,
+ "into": 9635,
+ "▁mail": 9636,
+ "▁authority": 9637,
+ "▁unable": 9638,
+ "filename": 9639,
+ "ék": 9640,
+ "лей": 9641,
+ "▁sector": 9642,
+ "appoint": 9643,
+ "▁hang": 9644,
+ "▁cel": 9645,
+ "related": 9646,
+ "itate": 9647,
+ "▁'<": 9648,
+ "amber": 9649,
+ "▁cheap": 9650,
+ "▁enabled": 9651,
+ "▁division": 9652,
+ "Any": 9653,
+ "▁hier": 9654,
+ "▁Head": 9655,
+ "ntax": 9656,
+ "uda": 9657,
+ "▁limitations": 9658,
+ "▁studio": 9659,
+ "media": 9660,
+ "▁circle": 9661,
+ "нова": 9662,
+ "▁laug": 9663,
+ "acts": 9664,
+ "▁Во": 9665,
+ "ód": 9666,
+ "pled": 9667,
+ "LOC": 9668,
+ "Expr": 9669,
+ ">:": 9670,
+ "▁prés": 9671,
+ "▁laughed": 9672,
+ "▁Three": 9673,
+ "лы": 9674,
+ "▁ends": 9675,
+ "▁fundament": 9676,
+ "▁inher": 9677,
+ "▁liv": 9678,
+ "bid": 9679,
+ "▁responsibility": 9680,
+ "▁checked": 9681,
+ "▁Pac": 9682,
+ "▁fault": 9683,
+ "▁yellow": 9684,
+ "▁salt": 9685,
+ "▁Francisco": 9686,
+ "▁^": 9687,
+ "▁ON": 9688,
+ "▁beauty": 9689,
+ "yg": 9690,
+ "▁Aff": 9691,
+ "▁Eq": 9692,
+ "▁magic": 9693,
+ "▁handler": 9694,
+ "xE": 9695,
+ "▁numerous": 9696,
+ "▁hole": 9697,
+ "▁rooms": 9698,
+ "cción": 9699,
+ "▁Arm": 9700,
+ "person": 9701,
+ "▁buildings": 9702,
+ "▁plate": 9703,
+ "bled": 9704,
+ "errors": 9705,
+ "▁Again": 9706,
+ "▁Default": 9707,
+ "▁Hard": 9708,
+ "tó": 9709,
+ "hus": 9710,
+ "▁dimension": 9711,
+ "iale": 9712,
+ "▁Mult": 9713,
+ "▁Government": 9714,
+ "Func": 9715,
+ "▁blow": 9716,
+ "▁rect": 9717,
+ "erra": 9718,
+ "connection": 9719,
+ "▁passing": 9720,
+ "ßen": 9721,
+ "phas": 9722,
+ "ensional": 9723,
+ "record": 9724,
+ "cohol": 9725,
+ "▁Harry": 9726,
+ "izontal": 9727,
+ "▁finger": 9728,
+ "▁younger": 9729,
+ "▁SC": 9730,
+ "operation": 9731,
+ "BY": 9732,
+ "heim": 9733,
+ "▁Bad": 9734,
+ "▁storm": 9735,
+ "▁Nat": 9736,
+ "▁buying": 9737,
+ "▁Sometimes": 9738,
+ "▁Ста": 9739,
+ "essed": 9740,
+ "▁damn": 9741,
+ "▁meg": 9742,
+ "umes": 9743,
+ "ünd": 9744,
+ "тра": 9745,
+ "▁silver": 9746,
+ "wd": 9747,
+ "hidden": 9748,
+ "ardo": 9749,
+ "▁communities": 9750,
+ "▁diet": 9751,
+ "otted": 9752,
+ "▁bat": 9753,
+ "ancer": 9754,
+ "▁fmt": 9755,
+ "▁Pen": 9756,
+ "▁til": 9757,
+ "Enum": 9758,
+ "PATH": 9759,
+ "▁matters": 9760,
+ "timeout": 9761,
+ "------------": 9762,
+ "kan": 9763,
+ "▁Corpor": 9764,
+ "=\"../../": 9765,
+ "▁Ale": 9766,
+ "hentication": 9767,
+ "▁complic": 9768,
+ "▁Security": 9769,
+ "OFF": 9770,
+ "Rad": 9771,
+ "apse": 9772,
+ "▁dance": 9773,
+ "▁permissions": 9774,
+ "▁warrant": 9775,
+ "▁lad": 9776,
+ "▁isol": 9777,
+ "dl": 9778,
+ "▁Au": 9779,
+ "yes": 9780,
+ "▁tv": 9781,
+ "▁provider": 9782,
+ "▁terrible": 9783,
+ "▁department": 9784,
+ "eral": 9785,
+ "▁implementation": 9786,
+ "SR": 9787,
+ "▁hearing": 9788,
+ "▁Kn": 9789,
+ "FR": 9790,
+ "tv": 9791,
+ "▁diss": 9792,
+ "FUN": 9793,
+ "▁durante": 9794,
+ "osis": 9795,
+ "▁tasks": 9796,
+ "▁Blo": 9797,
+ "вод": 9798,
+ "▁branch": 9799,
+ "▁politics": 9800,
+ "▁Elle": 9801,
+ "▁leadership": 9802,
+ "expr": 9803,
+ "▁techniques": 9804,
+ "prec": 9805,
+ "Sigma": 9806,
+ "imately": 9807,
+ "tk": 9808,
+ "achment": 9809,
+ "▁Enter": 9810,
+ "▁creative": 9811,
+ "▁зна": 9812,
+ "appy": 9813,
+ "unched": 9814,
+ "▁'',": 9815,
+ "onder": 9816,
+ "{-": 9817,
+ "NUM": 9818,
+ "▁narr": 9819,
+ "Memory": 9820,
+ "▁winning": 9821,
+ "▁Follow": 9822,
+ "*/\r": 9823,
+ "vision": 9824,
+ "resents": 9825,
+ "zione": 9826,
+ "▁latter": 9827,
+ "▁requests": 9828,
+ "▁margin": 9829,
+ "▁{\"": 9830,
+ "video": 9831,
+ "cn": 9832,
+ "▁Image": 9833,
+ "Tim": 9834,
+ "CONFIG": 9835,
+ "▁allowing": 9836,
+ "▁combined": 9837,
+ "PUT": 9838,
+ "▁instanceof": 9839,
+ "igin": 9840,
+ "▁pero": 9841,
+ "▁''": 9842,
+ "▁confidence": 9843,
+ "▁equivalent": 9844,
+ "pad": 9845,
+ "effect": 9846,
+ "RX": 9847,
+ "▁lang": 9848,
+ "strong": 9849,
+ "▁bridge": 9850,
+ "aya": 9851,
+ "▁treated": 9852,
+ "▁forth": 9853,
+ "SW": 9854,
+ "▁accounts": 9855,
+ "▁PO": 9856,
+ "▁listening": 9857,
+ "Route": 9858,
+ "()))": 9859,
+ "cpy": 9860,
+ "▁reform": 9861,
+ "▁gate": 9862,
+ "▁Walk": 9863,
+ "▁somehow": 9864,
+ "tf": 9865,
+ "▁layout": 9866,
+ "umin": 9867,
+ "▁considering": 9868,
+ "▁premi": 9869,
+ "▁Mom": 9870,
+ "athan": 9871,
+ "Gen": 9872,
+ "▁planet": 9873,
+ "amples": 9874,
+ "▁MO": 9875,
+ "shop": 9876,
+ "▁premier": 9877,
+ "▁simpl": 9878,
+ "▁segu": 9879,
+ "LY": 9880,
+ "Sum": 9881,
+ "▁tables": 9882,
+ "ska": 9883,
+ "▁ž": 9884,
+ "pd": 9885,
+ "▁sous": 9886,
+ "▁conference": 9887,
+ "▁Dat": 9888,
+ "Scroll": 9889,
+ "▁standards": 9890,
+ "▁гру": 9891,
+ "esse": 9892,
+ "▁citizens": 9893,
+ "▁occurred": 9894,
+ "▁democr": 9895,
+ "▁elev": 9896,
+ "▁Sem": 9897,
+ "ensus": 9898,
+ "headers": 9899,
+ "▁Chris": 9900,
+ "imento": 9901,
+ "kom": 9902,
+ "Cor": 9903,
+ "MIN": 9904,
+ "usher": 9905,
+ "Database": 9906,
+ "▁formal": 9907,
+ "igne": 9908,
+ "▁organizations": 9909,
+ "▁Ire": 9910,
+ "Xml": 9911,
+ "из": 9912,
+ "▁pray": 9913,
+ "▁bomb": 9914,
+ "▁mand": 9915,
+ "erts": 9916,
+ "▁clock": 9917,
+ "▁buck": 9918,
+ "вали": 9919,
+ "ensch": 9920,
+ "▁volt": 9921,
+ "▁films": 9922,
+ "▁plants": 9923,
+ "inode": 9924,
+ "Boolean": 9925,
+ "▁restaurant": 9926,
+ "ían": 9927,
+ "▁debut": 9928,
+ "pages": 9929,
+ "▁wordt": 9930,
+ "▁Ба": 9931,
+ "▁greatest": 9932,
+ "(\"/": 9933,
+ "▁copyright": 9934,
+ "▁rit": 9935,
+ "sizeof": 9936,
+ "Trace": 9937,
+ "uent": 9938,
+ "тур": 9939,
+ "▁ko": 9940,
+ ":\\": 9941,
+ "▁bigger": 9942,
+ "▁perfectly": 9943,
+ "tenance": 9944,
+ "MASK": 9945,
+ "ré": 9946,
+ "▁ett": 9947,
+ "▁nose": 9948,
+ "▁craft": 9949,
+ "iteral": 9950,
+ "▁discussed": 9951,
+ "▁Jewish": 9952,
+ "Cap": 9953,
+ "▁Unless": 9954,
+ "▁Jackson": 9955,
+ "Attributes": 9956,
+ "▁lunch": 9957,
+ "öl": 9958,
+ "atr": 9959,
+ "▁paying": 9960,
+ "Parse": 9961,
+ "()\r": 9962,
+ "lad": 9963,
+ "▁rare": 9964,
+ "▁[];": 9965,
+ "stone": 9966,
+ "▁unc": 9967,
+ "▁defense": 9968,
+ "}+": 9969,
+ "▁Global": 9970,
+ "▁Soviet": 9971,
+ "▁Australian": 9972,
+ "▁gli": 9973,
+ "variant": 9974,
+ "▁Ron": 9975,
+ "▁loan": 9976,
+ "Step": 9977,
+ "member": 9978,
+ "Sch": 9979,
+ "▁Committee": 9980,
+ "▁spending": 9981,
+ "▁Tri": 9982,
+ "▁Journal": 9983,
+ "▁sugar": 9984,
+ "elly": 9985,
+ "HTML": 9986,
+ "▁advent": 9987,
+ "wing": 9988,
+ "▁Whether": 9989,
+ "oration": 9990,
+ "▁NE": 9991,
+ "iveness": 9992,
+ "▁hav": 9993,
+ "▁conscious": 9994,
+ "een": 9995,
+ "Symbol": 9996,
+ "▁ку": 9997,
+ "Logger": 9998,
+ "▁Little": 9999,
+ "widet": 10000,
+ "ocation": 10001,
+ "pin": 10002,
+ "▁symmet": 10003,
+ "▁AD": 10004,
+ "▁posts": 10005,
+ "shal": 10006,
+ "▁Conf": 10007,
+ "▁chose": 10008,
+ "mal": 10009,
+ "ulo": 10010,
+ "▁Method": 10011,
+ "▁missed": 10012,
+ "Remove": 10013,
+ "Auto": 10014,
+ "VALUE": 10015,
+ "thlet": 10016,
+ "▁Force": 10017,
+ "pf": 10018,
+ "▁Я": 10019,
+ "late": 10020,
+ "▁pul": 10021,
+ "Pop": 10022,
+ "▁advanced": 10023,
+ "aires": 10024,
+ "ressed": 10025,
+ "AME": 10026,
+ "bell": 10027,
+ "aching": 10028,
+ "ić": 10029,
+ "echo": 10030,
+ "HS": 10031,
+ "▁funny": 10032,
+ "рии": 10033,
+ "▁eer": 10034,
+ "▁veget": 10035,
+ "▁fourth": 10036,
+ "cf": 10037,
+ "transform": 10038,
+ "▁grown": 10039,
+ "▁McC": 10040,
+ "site": 10041,
+ "▁beneath": 10042,
+ "▁shell": 10043,
+ "xd": 10044,
+ "Play": 10045,
+ "short": 10046,
+ "Role": 10047,
+ "▁religion": 10048,
+ "inator": 10049,
+ "}": 10050,
+ "▁Eliz": 10051,
+ "Microsoft": 10052,
+ "▁vez": 10053,
+ "▁рабо": 10054,
+ "reich": 10055,
+ "vet": 10056,
+ "enum": 10057,
+ "▁welcome": 10058,
+ "nament": 10059,
+ "▁jan": 10060,
+ "▁cycle": 10061,
+ "▁acknow": 10062,
+ "▁wound": 10063,
+ "idi": 10064,
+ "▁possibility": 10065,
+ "annotation": 10066,
+ "▁technical": 10067,
+ "▁fold": 10068,
+ "eh": 10069,
+ "istence": 10070,
+ "▁reply": 10071,
+ "etes": 10072,
+ "▁decades": 10073,
+ "wan": 10074,
+ "▁кра": 10075,
+ "▁Lab": 10076,
+ "▁unf": 10077,
+ "▁imper": 10078,
+ "▁bug": 10079,
+ "▁Though": 10080,
+ "throws": 10081,
+ "Visible": 10082,
+ "prev": 10083,
+ "▁Ty": 10084,
+ "▁depending": 10085,
+ "▁policies": 10086,
+ "andy": 10087,
+ "▁Italian": 10088,
+ "uma": 10089,
+ "▁signs": 10090,
+ "▁Through": 10091,
+ "бы": 10092,
+ "bot": 10093,
+ "▁publish": 10094,
+ ")**": 10095,
+ "ATTR": 10096,
+ "iral": 10097,
+ "VT": 10098,
+ "▁recognized": 10099,
+ "▁Lind": 10100,
+ "ection": 10101,
+ "▁relatively": 10102,
+ "▁Ah": 10103,
+ "▁Dig": 10104,
+ "ць": 10105,
+ "icket": 10106,
+ "▁specifically": 10107,
+ "nost": 10108,
+ "▁grass": 10109,
+ "▁causes": 10110,
+ "тво": 10111,
+ "utter": 10112,
+ "▁Festival": 10113,
+ "greg": 10114,
+ "▁weapons": 10115,
+ "▁sir": 10116,
+ "▁Virginia": 10117,
+ "login": 10118,
+ "▁schedul": 10119,
+ "ського": 10120,
+ "▁losing": 10121,
+ "▁Europ": 10122,
+ "\"><": 10123,
+ "asp": 10124,
+ "ajo": 10125,
+ "exports": 10126,
+ "▁Node": 10127,
+ "▁jako": 10128,
+ "▁ya": 10129,
+ "▁successfully": 10130,
+ "▁friendly": 10131,
+ "buff": 10132,
+ "DEFAULT": 10133,
+ "▁pregn": 10134,
+ "Required": 10135,
+ "▁binary": 10136,
+ "isting": 10137,
+ "▁stared": 10138,
+ "▁circumstances": 10139,
+ "▁хо": 10140,
+ "rei": 10141,
+ "▁Го": 10142,
+ "Transform": 10143,
+ "cnt": 10144,
+ "▁Ext": 10145,
+ "report": 10146,
+ "VERSION": 10147,
+ "▁analy": 10148,
+ "▁Marg": 10149,
+ "▁alleg": 10150,
+ "builder": 10151,
+ "ToString": 10152,
+ "Layer": 10153,
+ "íst": 10154,
+ "Prop": 10155,
+ "▁Emp": 10156,
+ "}]": 10157,
+ "▁selling": 10158,
+ "▁queue": 10159,
+ "▁seriously": 10160,
+ "▁Lead": 10161,
+ "textit": 10162,
+ "testing": 10163,
+ "▁Пре": 10164,
+ "security": 10165,
+ "iał": 10166,
+ "ún": 10167,
+ "chip": 10168,
+ "▁candidate": 10169,
+ "▁minister": 10170,
+ "eria": 10171,
+ "▁Het": 10172,
+ "дин": 10173,
+ "▁Britain": 10174,
+ "▁barely": 10175,
+ "▁sty": 10176,
+ "▁Spanish": 10177,
+ "▁Ven": 10178,
+ "timer": 10179,
+ "ків": 10180,
+ "▁documents": 10181,
+ "('.": 10182,
+ "▁debug": 10183,
+ "▁contro": 10184,
+ "стоя": 10185,
+ "▁joy": 10186,
+ "Sn": 10187,
+ "Inv": 10188,
+ "▁protocol": 10189,
+ "▁faces": 10190,
+ "▁Despite": 10191,
+ "sed": 10192,
+ "Conf": 10193,
+ "ARG": 10194,
+ "▁evolution": 10195,
+ "▁tod": 10196,
+ "▁Promise": 10197,
+ "▁posted": 10198,
+ "Perm": 10199,
+ "bet": 10200,
+ "Ang": 10201,
+ "Just": 10202,
+ "▁rum": 10203,
+ "layer": 10204,
+ "▁behavi": 10205,
+ "ipping": 10206,
+ "▁dynam": 10207,
+ "▁scheme": 10208,
+ "▁proto": 10209,
+ ")/": 10210,
+ "Collections": 10211,
+ "riev": 10212,
+ "▁Click": 10213,
+ "▁uns": 10214,
+ "widetilde": 10215,
+ "▁remembered": 10216,
+ "гі": 10217,
+ "inates": 10218,
+ "▁incorpor": 10219,
+ "▁Description": 10220,
+ "▁prepare": 10221,
+ "▁Final": 10222,
+ "uation": 10223,
+ "▁Queen": 10224,
+ ">;": 10225,
+ "▁automatically": 10226,
+ "▁sharp": 10227,
+ "▁meat": 10228,
+ "ateur": 10229,
+ "astern": 10230,
+ "▁stuck": 10231,
+ "ASSERT": 10232,
+ "▁planned": 10233,
+ "dots": 10234,
+ "ookie": 10235,
+ "▁Histor": 10236,
+ "▁reviews": 10237,
+ "IMP": 10238,
+ "▁answered": 10239,
+ "Total": 10240,
+ "▁sau": 10241,
+ "▁Mexico": 10242,
+ "continue": 10243,
+ "▁Apple": 10244,
+ "likely": 10245,
+ "зва": 10246,
+ "users": 10247,
+ "▁identified": 10248,
+ "▁Lev": 10249,
+ "▁mol": 10250,
+ "▁Islam": 10251,
+ "▁committed": 10252,
+ "writ": 10253,
+ "бер": 10254,
+ "rift": 10255,
+ "▁interrupt": 10256,
+ "▁readonly": 10257,
+ "schema": 10258,
+ "Sm": 10259,
+ "Double": 10260,
+ "aza": 10261,
+ "▁Hal": 10262,
+ "Move": 10263,
+ "▁Series": 10264,
+ "inline": 10265,
+ "▁которы": 10266,
+ "soc": 10267,
+ "▁tent": 10268,
+ "▁amer": 10269,
+ "aki": 10270,
+ "▁lady": 10271,
+ "▁tired": 10272,
+ "ifi": 10273,
+ "▁même": 10274,
+ "ouver": 10275,
+ "▁aside": 10276,
+ "Did": 10277,
+ "',\r": 10278,
+ "▁bringing": 10279,
+ "Drawing": 10280,
+ "aro": 10281,
+ "▁Rh": 10282,
+ "▁Naz": 10283,
+ "esso": 10284,
+ "▁reaction": 10285,
+ "mitted": 10286,
+ "▁absolute": 10287,
+ "haust": 10288,
+ "(()": 10289,
+ "▁Task": 10290,
+ "ERS": 10291,
+ "▁^{": 10292,
+ "VD": 10293,
+ "▁tone": 10294,
+ "dist": 10295,
+ "vs": 10296,
+ "▁wheel": 10297,
+ "▁administration": 10298,
+ "▁interests": 10299,
+ "▁pointer": 10300,
+ "▁encounter": 10301,
+ "aver": 10302,
+ "▁nord": 10303,
+ "ket": 10304,
+ "▁beach": 10305,
+ "▁enjoyed": 10306,
+ "contains": 10307,
+ "▁append": 10308,
+ "Wait": 10309,
+ "▁squad": 10310,
+ "zel": 10311,
+ "▁medium": 10312,
+ "▁sending": 10313,
+ "▁Lady": 10314,
+ "ções": 10315,
+ "▁destination": 10316,
+ "nych": 10317,
+ "▁conflict": 10318,
+ "▁Ly": 10319,
+ "▁vul": 10320,
+ "▁basically": 10321,
+ "reated": 10322,
+ "black": 10323,
+ "ugins": 10324,
+ "▁calm": 10325,
+ "érie": 10326,
+ "har": 10327,
+ "лан": 10328,
+ "▁Се": 10329,
+ "watch": 10330,
+ "▁Put": 10331,
+ "▁dump": 10332,
+ "acher": 10333,
+ "scroll": 10334,
+ "▁claimed": 10335,
+ "▁Control": 10336,
+ "▁blind": 10337,
+ "enti": 10338,
+ "▁Keep": 10339,
+ "▁Development": 10340,
+ "images": 10341,
+ "▁tough": 10342,
+ "gebra": 10343,
+ "▁sept": 10344,
+ "hew": 10345,
+ "▁skill": 10346,
+ "▁Tay": 10347,
+ "▁któ": 10348,
+ "owner": 10349,
+ "pare": 10350,
+ "▁fee": 10351,
+ "▁continues": 10352,
+ "▁kan": 10353,
+ "bes": 10354,
+ "▁cha": 10355,
+ "ovo": 10356,
+ "▁Night": 10357,
+ "icture": 10358,
+ "shire": 10359,
+ "▁essay": 10360,
+ "▁suppose": 10361,
+ "etic": 10362,
+ "Art": 10363,
+ "acon": 10364,
+ "lla": 10365,
+ "words": 10366,
+ "▁comparison": 10367,
+ "▁BE": 10368,
+ "▁challenges": 10369,
+ "▁ol": 10370,
+ "citep": 10371,
+ "▁Foot": 10372,
+ "▁Such": 10373,
+ "▁papers": 10374,
+ "activ": 10375,
+ "quer": 10376,
+ "тя": 10377,
+ "▁То": 10378,
+ "ський": 10379,
+ "thur": 10380,
+ "done": 10381,
+ "▁shock": 10382,
+ "▁dedicated": 10383,
+ "▁correspond": 10384,
+ "Second": 10385,
+ "▁bull": 10386,
+ "life": 10387,
+ "indent": 10388,
+ "▁figures": 10389,
+ "▁Andrew": 10390,
+ "isp": 10391,
+ "▁favour": 10392,
+ "зда": 10393,
+ "▁Elect": 10394,
+ "Full": 10395,
+ "▁nearby": 10396,
+ "▁Register": 10397,
+ "Scale": 10398,
+ "ications": 10399,
+ "ин": 10400,
+ "▁AM": 10401,
+ "pair": 10402,
+ "▁perspective": 10403,
+ "▁nos": 10404,
+ "apa": 10405,
+ "ostał": 10406,
+ "▁Pers": 10407,
+ "icer": 10408,
+ "▁plastic": 10409,
+ "дов": 10410,
+ "ciples": 10411,
+ "zą": 10412,
+ "clos": 10413,
+ "▁уча": 10414,
+ "▁Á": 10415,
+ "plugin": 10416,
+ "▁angle": 10417,
+ "▁commission": 10418,
+ "▁funds": 10419,
+ "▁indu": 10420,
+ "▁drawn": 10421,
+ "ám": 10422,
+ "▁developing": 10423,
+ "▁segment": 10424,
+ "isme": 10425,
+ "scr": 10426,
+ "▁lies": 10427,
+ "▁IL": 10428,
+ "▁api": 10429,
+ "Extension": 10430,
+ "▁scal": 10431,
+ "install": 10432,
+ "▁Week": 10433,
+ "▁gentle": 10434,
+ "▁Canadian": 10435,
+ "▁dialog": 10436,
+ "▁articles": 10437,
+ "Theme": 10438,
+ "SM": 10439,
+ "▁Bul": 10440,
+ "▁leur": 10441,
+ "▁stom": 10442,
+ "Plugin": 10443,
+ "▁после": 10444,
+ "▁stead": 10445,
+ "▁ś": 10446,
+ "ipher": 10447,
+ "▁prze": 10448,
+ "▁draft": 10449,
+ "bottom": 10450,
+ "▁{};": 10451,
+ "▁stayed": 10452,
+ "feature": 10453,
+ "▁vot": 10454,
+ "▁fabric": 10455,
+ "ça": 10456,
+ "('#": 10457,
+ "rea": 10458,
+ "▁reput": 10459,
+ "▁Cir": 10460,
+ "▁AL": 10461,
+ "▁assertEquals": 10462,
+ "results": 10463,
+ "▁Cross": 10464,
+ "ursday": 10465,
+ "▁audio": 10466,
+ "▁gap": 10467,
+ "▁streets": 10468,
+ "▁scientific": 10469,
+ "platform": 10470,
+ "▁auss": 10471,
+ "▁Cro": 10472,
+ "▁partial": 10473,
+ "unc": 10474,
+ "▁choices": 10475,
+ "▁или": 10476,
+ "pred": 10477,
+ "▁heads": 10478,
+ "terday": 10479,
+ "▁Nick": 10480,
+ "▁weird": 10481,
+ "asant": 10482,
+ "▁represented": 10483,
+ "▁пи": 10484,
+ "DP": 10485,
+ "orders": 10486,
+ "clock": 10487,
+ "▁Ho": 10488,
+ "arters": 10489,
+ "Cmd": 10490,
+ "oga": 10491,
+ "Keys": 10492,
+ "Report": 10493,
+ "▁Vill": 10494,
+ "▁Mu": 10495,
+ "▁owned": 10496,
+ "SUCCESS": 10497,
+ "▁typeof": 10498,
+ "hdr": 10499,
+ "uable": 10500,
+ "▁neighborhood": 10501,
+ "▁AP": 10502,
+ "▁resulting": 10503,
+ "▁shadow": 10504,
+ "STRING": 10505,
+ "▁videos": 10506,
+ "лення": 10507,
+ "expect": 10508,
+ "▁Valley": 10509,
+ "▁goto": 10510,
+ "▁Sher": 10511,
+ "frastr": 10512,
+ "▁operating": 10513,
+ "▁это": 10514,
+ "▁Licensed": 10515,
+ "Variable": 10516,
+ "▁PR": 10517,
+ "▁Hans": 10518,
+ "clone": 10519,
+ "▁Gesch": 10520,
+ "▁Band": 10521,
+ "........": 10522,
+ "uing": 10523,
+ "▁hundreds": 10524,
+ "▁ок": 10525,
+ "▁emotional": 10526,
+ "▁Indust": 10527,
+ ")+": 10528,
+ "▁Egypt": 10529,
+ "▁franç": 10530,
+ "▁š": 10531,
+ "▁fasc": 10532,
+ "onto": 10533,
+ "▁Adam": 10534,
+ "▁laid": 10535,
+ "▁rig": 10536,
+ "▁detailed": 10537,
+ "▁implements": 10538,
+ "▁university": 10539,
+ "▁Hy": 10540,
+ "▁grid": 10541,
+ "▁regions": 10542,
+ "Stop": 10543,
+ "▁slot": 10544,
+ "▁angry": 10545,
+ "▁-=": 10546,
+ "▁waited": 10547,
+ "Vert": 10548,
+ "\":\"": 10549,
+ "▁elem": 10550,
+ "▁rég": 10551,
+ "owed": 10552,
+ "Member": 10553,
+ "▁ratio": 10554,
+ "isen": 10555,
+ "▁Lem": 10556,
+ "gery": 10557,
+ "▁cream": 10558,
+ "▁était": 10559,
+ "▁geb": 10560,
+ "unique": 10561,
+ "▁Deb": 10562,
+ "▁factory": 10563,
+ "że": 10564,
+ "dialog": 10565,
+ "▁Config": 10566,
+ "Sync": 10567,
+ "angers": 10568,
+ "▁governing": 10569,
+ "▁Hun": 10570,
+ "Space": 10571,
+ "▁jest": 10572,
+ "icious": 10573,
+ "▁emphas": 10574,
+ "umps": 10575,
+ "▁Esp": 10576,
+ "▁sul": 10577,
+ "▁historical": 10578,
+ "ija": 10579,
+ "▁lying": 10580,
+ "▁Steve": 10581,
+ "▁measures": 10582,
+ "osto": 10583,
+ "?”": 10584,
+ "▁pocket": 10585,
+ "▁Sat": 10586,
+ "▁pitch": 10587,
+ "▁natur": 10588,
+ "▁humans": 10589,
+ "▁Simon": 10590,
+ "adores": 10591,
+ "(\"\\": 10592,
+ "inking": 10593,
+ "▁expos": 10594,
+ "material": 10595,
+ "▁apparently": 10596,
+ "▁Camb": 10597,
+ "▁Box": 10598,
+ "▁spaces": 10599,
+ "exists": 10600,
+ "▁acting": 10601,
+ "ORY": 10602,
+ "зова": 10603,
+ "Good": 10604,
+ "ienne": 10605,
+ "▁Williams": 10606,
+ "▁fruit": 10607,
+ "iera": 10608,
+ "▁Lim": 10609,
+ "▁trait": 10610,
+ "▁artists": 10611,
+ "▁absor": 10612,
+ "rait": 10613,
+ "LOAD": 10614,
+ "▁movies": 10615,
+ "▁dynamic": 10616,
+ "asts": 10617,
+ "▁Integer": 10618,
+ "▁smoke": 10619,
+ "пі": 10620,
+ "angel": 10621,
+ ">(\"": 10622,
+ "▁instrument": 10623,
+ "▁fuel": 10624,
+ "ної": 10625,
+ "atalogue": 10626,
+ "▁serial": 10627,
+ "Files": 10628,
+ "▁bathroom": 10629,
+ "ilo": 10630,
+ "esto": 10631,
+ "▁pm": 10632,
+ "entials": 10633,
+ "▁Online": 10634,
+ "white": 10635,
+ "▁tips": 10636,
+ "▁capable": 10637,
+ "Fig": 10638,
+ "TV": 10639,
+ "▁он": 10640,
+ "ké": 10641,
+ "bitr": 10642,
+ "Mapping": 10643,
+ "▁tak": 10644,
+ "ющи": 10645,
+ "вля": 10646,
+ ")\",": 10647,
+ "▁Karl": 10648,
+ "▁Human": 10649,
+ "▁Pot": 10650,
+ "▁represents": 10651,
+ "▁consistent": 10652,
+ "_(": 10653,
+ "wen": 10654,
+ "▁Rose": 10655,
+ "law": 10656,
+ "▁FROM": 10657,
+ "▁begins": 10658,
+ "▁edit": 10659,
+ "▁mountain": 10660,
+ "▁chapter": 10661,
+ "▁wondered": 10662,
+ "▁industrial": 10663,
+ "▁Major": 10664,
+ "▁ges": 10665,
+ "▁directed": 10666,
+ "eros": 10667,
+ "▁Wild": 10668,
+ "liament": 10669,
+ "Book": 10670,
+ "username": 10671,
+ "hot": 10672,
+ "▁nam": 10673,
+ "▁league": 10674,
+ "bra": 10675,
+ "кон": 10676,
+ "▁Tal": 10677,
+ "▁Ва": 10678,
+ "▁exports": 10679,
+ "(@": 10680,
+ "▁sharing": 10681,
+ "▁Tro": 10682,
+ "ść": 10683,
+ "uesday": 10684,
+ "ylv": 10685,
+ "▁guitar": 10686,
+ "elen": 10687,
+ "Selection": 10688,
+ "▁confident": 10689,
+ "rypto": 10690,
+ "▁hors": 10691,
+ "editor": 10692,
+ "▁shoulders": 10693,
+ "getName": 10694,
+ "encing": 10695,
+ "SELECT": 10696,
+ "вши": 10697,
+ "▁kinds": 10698,
+ "▁Wel": 10699,
+ "▁purposes": 10700,
+ "Matrix": 10701,
+ "invalid": 10702,
+ "▁owners": 10703,
+ "▁Records": 10704,
+ "▁Process": 10705,
+ "▁chat": 10706,
+ "▁Dor": 10707,
+ "▁bin": 10708,
+ "redit": 10709,
+ "oire": 10710,
+ "▁Total": 10711,
+ "▁Family": 10712,
+ "ARY": 10713,
+ "▁bread": 10714,
+ "▁compre": 10715,
+ "▁shoes": 10716,
+ "▁raz": 10717,
+ "▁trace": 10718,
+ "nej": 10719,
+ "orted": 10720,
+ "hn": 10721,
+ "▁procedure": 10722,
+ "properties": 10723,
+ "plier": 10724,
+ "▁hero": 10725,
+ "panel": 10726,
+ "▁marked": 10727,
+ "▁worried": 10728,
+ "\\|": 10729,
+ "pts": 10730,
+ "▁Support": 10731,
+ "▁serving": 10732,
+ "Fail": 10733,
+ "▁disappoint": 10734,
+ "▁Scot": 10735,
+ "▁pleasure": 10736,
+ "▁judge": 10737,
+ "zeich": 10738,
+ "▁forever": 10739,
+ "▁Zeit": 10740,
+ "uous": 10741,
+ "inent": 10742,
+ "▁dw": 10743,
+ "▁waren": 10744,
+ "▁flash": 10745,
+ "▁troops": 10746,
+ "▁drugs": 10747,
+ "▁diam": 10748,
+ ".~": 10749,
+ "imp": 10750,
+ "inned": 10751,
+ "▁EV": 10752,
+ "Struct": 10753,
+ "▁justice": 10754,
+ "▁officials": 10755,
+ "ffff": 10756,
+ "▁Common": 10757,
+ "▁Cat": 10758,
+ "▁tomorrow": 10759,
+ "▁él": 10760,
+ "Texture": 10761,
+ "qpoint": 10762,
+ "▁Fried": 10763,
+ "▁Term": 10764,
+ "pgfqpoint": 10765,
+ "▁nem": 10766,
+ "norm": 10767,
+ "▁hardly": 10768,
+ "oda": 10769,
+ "zeta": 10770,
+ "emic": 10771,
+ "▁полу": 10772,
+ "▁loaded": 10773,
+ "kes": 10774,
+ "ció": 10775,
+ "▁fool": 10776,
+ "▁trick": 10777,
+ "▁dst": 10778,
+ "Find": 10779,
+ "▁все": 10780,
+ "}},": 10781,
+ "▁framework": 10782,
+ "▁merely": 10783,
+ "▁union": 10784,
+ "▁Edward": 10785,
+ "rif": 10786,
+ "Flag": 10787,
+ "▁crisis": 10788,
+ "▁finite": 10789,
+ "▁lol": 10790,
+ "▁Kim": 10791,
+ "ната": 10792,
+ "since": 10793,
+ "▁compat": 10794,
+ "▁pert": 10795,
+ "ibilities": 10796,
+ "▁también": 10797,
+ "ibli": 10798,
+ "▁teen": 10799,
+ "▁sympt": 10800,
+ "oral": 10801,
+ "ders": 10802,
+ "otte": 10803,
+ "при": 10804,
+ "▁Jane": 10805,
+ "▁originally": 10806,
+ "▁throat": 10807,
+ "mag": 10808,
+ "sup": 10809,
+ "uni": 10810,
+ "$$": 10811,
+ "▁Library": 10812,
+ "▁attacks": 10813,
+ "ingen": 10814,
+ "('/": 10815,
+ "▁hes": 10816,
+ "coin": 10817,
+ "ounce": 10818,
+ "▁Academy": 10819,
+ "MODULE": 10820,
+ "isms": 10821,
+ "▁Adv": 10822,
+ "▁Bol": 10823,
+ "▁incident": 10824,
+ ")^{": 10825,
+ "▁bij": 10826,
+ "▁Rome": 10827,
+ "▁Italy": 10828,
+ "events": 10829,
+ "▁Fern": 10830,
+ "▁ber": 10831,
+ "▁silent": 10832,
+ "▁pier": 10833,
+ "▁YO": 10834,
+ "▁plain": 10835,
+ "Bas": 10836,
+ "▁pill": 10837,
+ "rase": 10838,
+ "▁carrying": 10839,
+ "▁resp": 10840,
+ "ную": 10841,
+ "▁typical": 10842,
+ "Wrapper": 10843,
+ "▁gau": 10844,
+ "▁chemical": 10845,
+ "▁hal": 10846,
+ "throw": 10847,
+ "Cluster": 10848,
+ "▁Gab": 10849,
+ "▁Girl": 10850,
+ "quir": 10851,
+ "▁Arg": 10852,
+ "▁relief": 10853,
+ "▁Ве": 10854,
+ "dm": 10855,
+ "▁frustr": 10856,
+ "\\%": 10857,
+ "▁stores": 10858,
+ "▁bottle": 10859,
+ "▁Lew": 10860,
+ "two": 10861,
+ "stad": 10862,
+ "▁cheek": 10863,
+ "▁concerns": 10864,
+ "▁helpful": 10865,
+ "▁coverage": 10866,
+ "isi": 10867,
+ "ADD": 10868,
+ "async": 10869,
+ "▁approximately": 10870,
+ "iffer": 10871,
+ "hook": 10872,
+ "▁enum": 10873,
+ "ová": 10874,
+ "▁evil": 10875,
+ "▁constantly": 10876,
+ "apply": 10877,
+ "▁siè": 10878,
+ "▁practices": 10879,
+ "▁teachers": 10880,
+ "▁Sn": 10881,
+ "▁Awards": 10882,
+ "▁substant": 10883,
+ "▁$.": 10884,
+ "dk": 10885,
+ "▁mob": 10886,
+ "▁ingred": 10887,
+ "vere": 10888,
+ "Multi": 10889,
+ "пер": 10890,
+ "stal": 10891,
+ "yard": 10892,
+ "required": 10893,
+ "vement": 10894,
+ "▁intelligence": 10895,
+ "▁thinks": 10896,
+ "▁personally": 10897,
+ "▁trained": 10898,
+ "orney": 10899,
+ ")": 10900,
+ "gged": 10901,
+ "EINVAL": 10902,
+ "arna": 10903,
+ "▁Hamilton": 10904,
+ "merce": 10905,
+ "ekt": 10906,
+ "OF": 10907,
+ ")[": 10908,
+ "rug": 10909,
+ "ición": 10910,
+ "▁survey": 10911,
+ "nesday": 10912,
+ "▁pag": 10913,
+ "▁boundary": 10914,
+ "▁quantum": 10915,
+ "▁drawing": 10916,
+ "▁volunte": 10917,
+ "▁Word": 10918,
+ "sky": 10919,
+ "▁Greg": 10920,
+ "coll": 10921,
+ "hide": 10922,
+ "▁swim": 10923,
+ "▁revealed": 10924,
+ "adv": 10925,
+ "дя": 10926,
+ ".\");": 10927,
+ "▁explan": 10928,
+ "▁Current": 10929,
+ "▁gotten": 10930,
+ "▁falling": 10931,
+ "▁contained": 10932,
+ "UND": 10933,
+ "▁Should": 10934,
+ "▁killing": 10935,
+ "▁aspects": 10936,
+ "icted": 10937,
+ "▁Param": 10938,
+ "\",\r": 10939,
+ "TION": 10940,
+ "));\r": 10941,
+ "▁Iran": 10942,
+ "beit": 10943,
+ "▁Bu": 10944,
+ "▁[],": 10945,
+ "SSION": 10946,
+ "▁Mah": 10947,
+ "▁resolution": 10948,
+ "▁boss": 10949,
+ "lg": 10950,
+ "chor": 10951,
+ "▁Unter": 10952,
+ "▁debt": 10953,
+ "▁vid": 10954,
+ "gie": 10955,
+ "▁uno": 10956,
+ "CB": 10957,
+ "plom": 10958,
+ "LICENSE": 10959,
+ "▁Kenn": 10960,
+ "▁finns": 10961,
+ "ONG": 10962,
+ "▁somewhat": 10963,
+ "▁actor": 10964,
+ "▁Status": 10965,
+ "▁probability": 10966,
+ "fb": 10967,
+ "▁chart": 10968,
+ "▁stands": 10969,
+ "policy": 10970,
+ "▁onder": 10971,
+ "tabular": 10972,
+ "▁Ash": 10973,
+ "▁boost": 10974,
+ "▁desper": 10975,
+ "month": 10976,
+ "▁alert": 10977,
+ "▁suite": 10978,
+ "▁gén": 10979,
+ "▁vacc": 10980,
+ "▁Has": 10981,
+ "Mask": 10982,
+ "▁Thursday": 10983,
+ "▁proved": 10984,
+ "▁Nel": 10985,
+ "▁moral": 10986,
+ "▁ja": 10987,
+ "auer": 10988,
+ "codec": 10989,
+ "▁instant": 10990,
+ "amps": 10991,
+ "▁milk": 10992,
+ "WORD": 10993,
+ "▁Ö": 10994,
+ "Email": 10995,
+ "Elements": 10996,
+ "▁forma": 10997,
+ "Free": 10998,
+ "MAP": 10999,
+ "▁Ж": 11000,
+ "sym": 11001,
+ "▁ти": 11002,
+ "▁Econom": 11003,
+ "▁Vi": 11004,
+ "▁Columb": 11005,
+ "▁_,": 11006,
+ "oret": 11007,
+ "Sequ": 11008,
+ "plan": 11009,
+ "▁frequency": 11010,
+ "irement": 11011,
+ "▁assumed": 11012,
+ "▁Ca": 11013,
+ "▁Bit": 11014,
+ "▁коман": 11015,
+ "▁smell": 11016,
+ "Security": 11017,
+ "▁aqu": 11018,
+ "oor": 11019,
+ "price": 11020,
+ "inity": 11021,
+ "▁axis": 11022,
+ "release": 11023,
+ "▁resolve": 11024,
+ "▁tears": 11025,
+ "▁bother": 11026,
+ "▁Community": 11027,
+ "▁registered": 11028,
+ "▁revolution": 11029,
+ "?.": 11030,
+ "▁versions": 11031,
+ "%%%%": 11032,
+ "ydro": 11033,
+ "Success": 11034,
+ "▁Win": 11035,
+ "▁Boy": 11036,
+ "▁Dub": 11037,
+ "▁kw": 11038,
+ "▁noch": 11039,
+ "▁charges": 11040,
+ "arios": 11041,
+ "uar": 11042,
+ ";&": 11043,
+ "▁había": 11044,
+ "(`": 11045,
+ "▁tx": 11046,
+ "elve": 11047,
+ "▁años": 11048,
+ "▁math": 11049,
+ "▁Alf": 11050,
+ "▁Fund": 11051,
+ "▁manifest": 11052,
+ "▁attached": 11053,
+ "▁spiritual": 11054,
+ "▁Alexander": 11055,
+ "unes": 11056,
+ "▁seed": 11057,
+ "▁Но": 11058,
+ "▁magazine": 11059,
+ "▁eigen": 11060,
+ "▁обра": 11061,
+ "ea": 11062,
+ "▁PH": 11063,
+ "swing": 11064,
+ "▁Asia": 11065,
+ "ју": 11066,
+ "▁KIND": 11067,
+ "Identifier": 11068,
+ "once": 11069,
+ "▁alcohol": 11070,
+ "ції": 11071,
+ "styles": 11072,
+ "assertEqual": 11073,
+ "▁Ra": 11074,
+ "графи": 11075,
+ "▁millions": 11076,
+ "▁chunk": 11077,
+ "дер": 11078,
+ "Package": 11079,
+ "UST": 11080,
+ "▁Nothing": 11081,
+ "(\"#": 11082,
+ "▁Mid": 11083,
+ "▁нача": 11084,
+ "ły": 11085,
+ "AAAA": 11086,
+ "▁launched": 11087,
+ "▁wake": 11088,
+ "▁guests": 11089,
+ "▁differences": 11090,
+ "udi": 11091,
+ "▁aid": 11092,
+ "▁Sport": 11093,
+ "ulator": 11094,
+ "execute": 11095,
+ "plot": 11096,
+ "ching": 11097,
+ "▁Norm": 11098,
+ "tm": 11099,
+ "\\+": 11100,
+ "ARD": 11101,
+ "▁beer": 11102,
+ "▁під": 11103,
+ "IAL": 11104,
+ "storage": 11105,
+ "▁Anna": 11106,
+ "▁yards": 11107,
+ "▁technique": 11108,
+ "▁où": 11109,
+ "atten": 11110,
+ "UNT": 11111,
+ "don": 11112,
+ "фор": 11113,
+ "▁hoping": 11114,
+ "▁victory": 11115,
+ "itat": 11116,
+ "▁significantly": 11117,
+ "▁practical": 11118,
+ "ije": 11119,
+ "▁expansion": 11120,
+ "JS": 11121,
+ "ixels": 11122,
+ "USER": 11123,
+ "Shape": 11124,
+ "▁extent": 11125,
+ "lio": 11126,
+ "▁pued": 11127,
+ "olid": 11128,
+ "▁gam": 11129,
+ "▁sevent": 11130,
+ "▁Ga": 11131,
+ "anguages": 11132,
+ "(((": 11133,
+ "ъл": 11134,
+ "▁Exper": 11135,
+ "asty": 11136,
+ "rieg": 11137,
+ "gio": 11138,
+ "odo": 11139,
+ "▁colle": 11140,
+ "▁stored": 11141,
+ "▁Sche": 11142,
+ "istant": 11143,
+ "▁lip": 11144,
+ "BR": 11145,
+ "▁aug": 11146,
+ "▁Search": 11147,
+ ")=\\": 11148,
+ "▁Ur": 11149,
+ "▁sole": 11150,
+ "illo": 11151,
+ "▁mehr": 11152,
+ "kit": 11153,
+ "▁interior": 11154,
+ "LIST": 11155,
+ "adel": 11156,
+ "▁shopping": 11157,
+ "▁slä": 11158,
+ "Your": 11159,
+ "DITION": 11160,
+ "▁Http": 11161,
+ "raham": 11162,
+ "три": 11163,
+ "▁brings": 11164,
+ "Rev": 11165,
+ "▁propag": 11166,
+ "ityEngine": 11167,
+ "()),": 11168,
+ "▁ingår": 11169,
+ "▁Ireland": 11170,
+ "▁\"./": 11171,
+ "▁Harr": 11172,
+ "▁admin": 11173,
+ "eno": 11174,
+ "▁kr": 11175,
+ "▁está": 11176,
+ "▁props": 11177,
+ "tok": 11178,
+ "omorph": 11179,
+ "▁affected": 11180,
+ "Phone": 11181,
+ "▁degrees": 11182,
+ "some": 11183,
+ "▁nin": 11184,
+ "EVENT": 11185,
+ "▁interaction": 11186,
+ "▁Tuesday": 11187,
+ "iterator": 11188,
+ "▁Nob": 11189,
+ "▁scatter": 11190,
+ "ucket": 11191,
+ "complete": 11192,
+ "▁duty": 11193,
+ "▁answers": 11194,
+ "Progress": 11195,
+ "eed": 11196,
+ "рон": 11197,
+ "▁vie": 11198,
+ "▁depos": 11199,
+ "▁packet": 11200,
+ "▁tow": 11201,
+ "▁deleg": 11202,
+ "audio": 11203,
+ "▁vary": 11204,
+ "▁migr": 11205,
+ "фі": 11206,
+ "esa": 11207,
+ "Events": 11208,
+ "haus": 11209,
+ "▁Sav": 11210,
+ "▁Portug": 11211,
+ "▁сто": 11212,
+ "ilation": 11213,
+ "▁metadata": 11214,
+ "las": 11215,
+ "▁ai": 11216,
+ "▁anger": 11217,
+ "▁ham": 11218,
+ "▁Anal": 11219,
+ "▁frequently": 11220,
+ "▁FALSE": 11221,
+ "oche": 11222,
+ "rez": 11223,
+ "▁Viet": 11224,
+ "quis": 11225,
+ "▁charged": 11226,
+ "äs": 11227,
+ "▁Path": 11228,
+ "▁accurate": 11229,
+ "▁Plus": 11230,
+ "keit": 11231,
+ "▁Input": 11232,
+ "when": 11233,
+ "eras": 11234,
+ "▁воз": 11235,
+ "▁derived": 11236,
+ "aje": 11237,
+ "▁Had": 11238,
+ "uren": 11239,
+ "ór": 11240,
+ "}=\\": 11241,
+ "ureau": 11242,
+ "aland": 11243,
+ "Execution": 11244,
+ "eden": 11245,
+ "▁seeking": 11246,
+ "changed": 11247,
+ "▁trem": 11248,
+ "ску": 11249,
+ "▁Geme": 11250,
+ "inating": 11251,
+ "▁columns": 11252,
+ "EP": 11253,
+ "▁injury": 11254,
+ "endent": 11255,
+ "▁headed": 11256,
+ "ASE": 11257,
+ "▁Muslim": 11258,
+ "▁climate": 11259,
+ "▁fake": 11260,
+ "CMD": 11261,
+ "ји": 11262,
+ "▁Arts": 11263,
+ "fection": 11264,
+ "▁pit": 11265,
+ ">\\": 11266,
+ "anal": 11267,
+ "Section": 11268,
+ "plus": 11269,
+ "üt": 11270,
+ "▁embed": 11271,
+ "▁strings": 11272,
+ "Before": 11273,
+ "proc": 11274,
+ "▁спо": 11275,
+ "trl": 11276,
+ "vr": 11277,
+ "Background": 11278,
+ "logger": 11279,
+ "agraph": 11280,
+ "iest": 11281,
+ "▁goods": 11282,
+ "batch": 11283,
+ "▁optional": 11284,
+ "▁Taylor": 11285,
+ "▁recognize": 11286,
+ "walk": 11287,
+ "▁Hit": 11288,
+ "▁Elizabeth": 11289,
+ "}:": 11290,
+ "▁careful": 11291,
+ "краї": 11292,
+ "▁locations": 11293,
+ "▁structures": 11294,
+ "▁disk": 11295,
+ "▁ships": 11296,
+ "▁suo": 11297,
+ "▁sowie": 11298,
+ "▁Ess": 11299,
+ "▁Hash": 11300,
+ "▁reasonable": 11301,
+ "▁Moreover": 11302,
+ "▁formula": 11303,
+ "▁Centre": 11304,
+ "▁residents": 11305,
+ "RS": 11306,
+ "Ids": 11307,
+ "▁Know": 11308,
+ "▁trib": 11309,
+ "▁rés": 11310,
+ "▁stable": 11311,
+ "▁Would": 11312,
+ "▁breaking": 11313,
+ "▁meal": 11314,
+ "▁phen": 11315,
+ "▁fel": 11316,
+ "▁Fred": 11317,
+ "Author": 11318,
+ "▁capture": 11319,
+ "opts": 11320,
+ "▁everywhere": 11321,
+ "▁sque": 11322,
+ "▁moder": 11323,
+ "setup": 11324,
+ "▁Supp": 11325,
+ "▁whenever": 11326,
+ "{(": 11327,
+ "wart": 11328,
+ "▁toe": 11329,
+ "Prefix": 11330,
+ "hou": 11331,
+ "gage": 11332,
+ ">\"": 11333,
+ "▁frag": 11334,
+ "▁Theorem": 11335,
+ "memory": 11336,
+ "▁contents": 11337,
+ "docs": 11338,
+ "}'": 11339,
+ "▁Irish": 11340,
+ "Then": 11341,
+ "aats": 11342,
+ "Save": 11343,
+ "▁agency": 11344,
+ "▁име": 11345,
+ "дова": 11346,
+ "▁Function": 11347,
+ "NN": 11348,
+ "destroy": 11349,
+ "▁Message": 11350,
+ "▁cancel": 11351,
+ "▁superior": 11352,
+ "▁ec": 11353,
+ "▁literature": 11354,
+ "▁PART": 11355,
+ "Il": 11356,
+ "▁Cab": 11357,
+ "engine": 11358,
+ "▁basket": 11359,
+ "worth": 11360,
+ "▁Sel": 11361,
+ "fetch": 11362,
+ "▁Stadt": 11363,
+ "▁Ки": 11364,
+ "▁conj": 11365,
+ "▁seiner": 11366,
+ "▁confirmed": 11367,
+ "▁Argent": 11368,
+ "amar": 11369,
+ "pgfpath": 11370,
+ "▁struggle": 11371,
+ "Pattern": 11372,
+ "▁Middle": 11373,
+ "itan": 11374,
+ "▁moon": 11375,
+ "orough": 11376,
+ "▁Catholic": 11377,
+ "▁struck": 11378,
+ "]->": 11379,
+ "▁weapon": 11380,
+ "▁subst": 11381,
+ "▁instructions": 11382,
+ "▁occas": 11383,
+ "protected": 11384,
+ "▁Less": 11385,
+ "▁batch": 11386,
+ "▁contra": 11387,
+ "▁deck": 11388,
+ "▁ignored": 11389,
+ "▁refused": 11390,
+ "trigger": 11391,
+ "▁criminal": 11392,
+ "GA": 11393,
+ "olly": 11394,
+ "▁Bell": 11395,
+ "▁Ю": 11396,
+ "forward": 11397,
+ "▁prefix": 11398,
+ "▁immediate": 11399,
+ "▁assigned": 11400,
+ "▁elected": 11401,
+ "▁tonight": 11402,
+ "▁Dies": 11403,
+ "▁Beach": 11404,
+ "▁preced": 11405,
+ "ował": 11406,
+ "▁galax": 11407,
+ "▁logic": 11408,
+ "enza": 11409,
+ "▁Captain": 11410,
+ "▁Hay": 11411,
+ "▁facts": 11412,
+ "▁ни": 11413,
+ "té": 11414,
+ "▁sb": 11415,
+ "oped": 11416,
+ "▁combat": 11417,
+ "▁explore": 11418,
+ "▁(-": 11419,
+ "Loader": 11420,
+ "▁Wilson": 11421,
+ "▁locked": 11422,
+ ":": 11423,
+ "▁Od": 11424,
+ "▁Prote": 11425,
+ "▁disabled": 11426,
+ "▁hatte": 11427,
+ "▁shout": 11428,
+ "▁constructor": 11429,
+ "бі": 11430,
+ "▁tras": 11431,
+ "▁Father": 11432,
+ "▁adj": 11433,
+ "▁Carolina": 11434,
+ "▁Food": 11435,
+ "bad": 11436,
+ "atore": 11437,
+ "parameters": 11438,
+ "▁Full": 11439,
+ "[-": 11440,
+ "▁\"#": 11441,
+ "▁Try": 11442,
+ "ської": 11443,
+ "▁exhaust": 11444,
+ "▁scroll": 11445,
+ "_;": 11446,
+ "Who": 11447,
+ "▁delivered": 11448,
+ "▁referred": 11449,
+ "▁prospect": 11450,
+ "scan": 11451,
+ "▁modified": 11452,
+ "Generator": 11453,
+ "▁excess": 11454,
+ "▁kg": 11455,
+ "zet": 11456,
+ "icz": 11457,
+ "clipse": 11458,
+ "▁tank": 11459,
+ "▁guns": 11460,
+ "▁Ges": 11461,
+ "inton": 11462,
+ "▁Wednesday": 11463,
+ "▁mainly": 11464,
+ "parser": 11465,
+ "▁effectively": 11466,
+ "▁Ку": 11467,
+ "▁resident": 11468,
+ "▁Li": 11469,
+ "▁flying": 11470,
+ "▁mayor": 11471,
+ "üh": 11472,
+ "uta": 11473,
+ "▁colour": 11474,
+ "▁aircraft": 11475,
+ "terior": 11476,
+ "nr": 11477,
+ "▁keeps": 11478,
+ "fan": 11479,
+ "▁shirt": 11480,
+ "Compar": 11481,
+ "▁Eth": 11482,
+ "Mac": 11483,
+ "clean": 11484,
+ "slice": 11485,
+ "czy": 11486,
+ "▁gender": 11487,
+ "▁butter": 11488,
+ "AUT": 11489,
+ "▁Element": 11490,
+ "Fin": 11491,
+ "dma": 11492,
+ "sample": 11493,
+ "Registry": 11494,
+ "▁classic": 11495,
+ "▁drove": 11496,
+ "pb": 11497,
+ "defined": 11498,
+ "▁reward": 11499,
+ "yal": 11500,
+ "]),": 11501,
+ "▁BAS": 11502,
+ "▁hyper": 11503,
+ "▁Ни": 11504,
+ "▁).": 11505,
+ "Psi": 11506,
+ "▁entries": 11507,
+ "▁Kingdom": 11508,
+ "▁Song": 11509,
+ "▁prompt": 11510,
+ "centering": 11511,
+ "▁Holly": 11512,
+ "eman": 11513,
+ "▁painting": 11514,
+ "▁formation": 11515,
+ "▁Request": 11516,
+ "controller": 11517,
+ "Region": 11518,
+ "PY": 11519,
+ "idades": 11520,
+ "TL": 11521,
+ "▁disable": 11522,
+ "▁rein": 11523,
+ "rical": 11524,
+ "\"\r": 11525,
+ "%)": 11526,
+ "▁Sab": 11527,
+ "▁Without": 11528,
+ "Serv": 11529,
+ "▁Short": 11530,
+ "▁ю": 11531,
+ "▁resc": 11532,
+ "▁patterns": 11533,
+ "▁ArrayList": 11534,
+ "symbol": 11535,
+ "aco": 11536,
+ "▁Hom": 11537,
+ "help": 11538,
+ "▁hasta": 11539,
+ "▁installed": 11540,
+ "atie": 11541,
+ "▁visited": 11542,
+ "▁Бе": 11543,
+ "){\\": 11544,
+ "▁desde": 11545,
+ "JECT": 11546,
+ "▁drew": 11547,
+ "▁Stock": 11548,
+ "▁Cru": 11549,
+ "DEF": 11550,
+ "obby": 11551,
+ "izable": 11552,
+ "ogether": 11553,
+ "▁aber": 11554,
+ "▁dan": 11555,
+ "alis": 11556,
+ "tail": 11557,
+ "▁expressed": 11558,
+ "▁Access": 11559,
+ "Seg": 11560,
+ "▁Lib": 11561,
+ "▁supports": 11562,
+ "background": 11563,
+ "▁commune": 11564,
+ "called": 11565,
+ "▁printf": 11566,
+ "▁Prince": 11567,
+ "ните": 11568,
+ "depend": 11569,
+ "▁dels": 11570,
+ "neur": 11571,
+ "▁recommended": 11572,
+ "▁founded": 11573,
+ "▁markets": 11574,
+ "▁destroyed": 11575,
+ "▁abstract": 11576,
+ "▁serie": 11577,
+ "▁Dun": 11578,
+ "Term": 11579,
+ "▁portion": 11580,
+ "adapter": 11581,
+ "isset": 11582,
+ "чески": 11583,
+ "▁integer": 11584,
+ "▁returning": 11585,
+ "enties": 11586,
+ "▁Fair": 11587,
+ "▁USB": 11588,
+ "▁Price": 11589,
+ "igate": 11590,
+ "▁settled": 11591,
+ "({\\": 11592,
+ "nek": 11593,
+ "▁therm": 11594,
+ "▁cig": 11595,
+ "ány": 11596,
+ "▁investigation": 11597,
+ "ometer": 11598,
+ "SUP": 11599,
+ "Some": 11600,
+ "sing": 11601,
+ "Constant": 11602,
+ "▁retail": 11603,
+ "ży": 11604,
+ "▁drinking": 11605,
+ "▁Invest": 11606,
+ "SV": 11607,
+ "iginal": 11608,
+ "▁Bow": 11609,
+ "{{\\": 11610,
+ "▁assistance": 11611,
+ "▁intellect": 11612,
+ "INIT": 11613,
+ "aug": 11614,
+ "▁Leon": 11615,
+ "Sur": 11616,
+ "▁admit": 11617,
+ "▁Command": 11618,
+ "illes": 11619,
+ "rov": 11620,
+ "▁oh": 11621,
+ "▁não": 11622,
+ "▁matching": 11623,
+ "▁genu": 11624,
+ "▁Ox": 11625,
+ "тся": 11626,
+ "notation": 11627,
+ "GO": 11628,
+ "▁Nap": 11629,
+ "▁verify": 11630,
+ "▁aussi": 11631,
+ "DateTime": 11632,
+ "▁suitable": 11633,
+ "▁indicate": 11634,
+ "▁Live": 11635,
+ "Feature": 11636,
+ "▁tracks": 11637,
+ "▁hasn": 11638,
+ "▁Java": 11639,
+ "▁closely": 11640,
+ "▁Dad": 11641,
+ "ceive": 11642,
+ "▁Market": 11643,
+ "agy": 11644,
+ "▁\"-": 11645,
+ "awn": 11646,
+ "stell": 11647,
+ "pton": 11648,
+ "zeit": 11649,
+ "▁Vector": 11650,
+ "▁MAX": 11651,
+ "▁Federal": 11652,
+ "wall": 11653,
+ "▁Jen": 11654,
+ "delay": 11655,
+ "▁limits": 11656,
+ "▁Quest": 11657,
+ "Cam": 11658,
+ "▁Fel": 11659,
+ "writer": 11660,
+ "LP": 11661,
+ "▁moves": 11662,
+ "▁Execut": 11663,
+ "▁DB": 11664,
+ "oker": 11665,
+ "scribe": 11666,
+ "elijk": 11667,
+ "Constants": 11668,
+ "Addr": 11669,
+ "▁}}": 11670,
+ "▁channels": 11671,
+ "iy": 11672,
+ "riority": 11673,
+ "▁trading": 11674,
+ "▁facilities": 11675,
+ "▁Pack": 11676,
+ "▁sys": 11677,
+ "▁meta": 11678,
+ "▁estimate": 11679,
+ "▁Later": 11680,
+ "issue": 11681,
+ "▁Having": 11682,
+ "▁guest": 11683,
+ "▁nobody": 11684,
+ "depth": 11685,
+ "▁został": 11686,
+ "пера": 11687,
+ ")}\\": 11688,
+ "bg": 11689,
+ "▁Twitter": 11690,
+ "▁darkness": 11691,
+ "jpg": 11692,
+ "contr": 11693,
+ "kernel": 11694,
+ "]\\": 11695,
+ "▁extend": 11696,
+ "roc": 11697,
+ "NET": 11698,
+ "MSG": 11699,
+ "▁burst": 11700,
+ "▁repair": 11701,
+ "▁fetch": 11702,
+ "ieg": 11703,
+ "ús": 11704,
+ "Screen": 11705,
+ "blem": 11706,
+ "AppCompat": 11707,
+ "▁chap": 11708,
+ "ELD": 11709,
+ "▁Penn": 11710,
+ "▁promote": 11711,
+ "▁Ukr": 11712,
+ "arest": 11713,
+ "▁samples": 11714,
+ "▁Greek": 11715,
+ "▁constru": 11716,
+ "▁universe": 11717,
+ "elijke": 11718,
+ "▁preferred": 11719,
+ "▁Де": 11720,
+ "▁Ira": 11721,
+ "▁dow": 11722,
+ "agues": 11723,
+ "HERE": 11724,
+ "▁experts": 11725,
+ "Protocol": 11726,
+ "PIO": 11727,
+ "▁naz": 11728,
+ "▁Kh": 11729,
+ "hör": 11730,
+ "▁distingu": 11731,
+ "▁BY": 11732,
+ "▁seine": 11733,
+ "eping": 11734,
+ "▁fairly": 11735,
+ "▁Mean": 11736,
+ "ixer": 11737,
+ "insi": 11738,
+ "▁authors": 11739,
+ "**.": 11740,
+ "AI": 11741,
+ "▁edges": 11742,
+ "▁shooting": 11743,
+ "Admin": 11744,
+ "▁maps": 11745,
+ "chant": 11746,
+ "▁COVID": 11747,
+ "▁linked": 11748,
+ "▁ske": 11749,
+ "▁powers": 11750,
+ "ád": 11751,
+ "▁stomach": 11752,
+ "▁usage": 11753,
+ "▁defend": 11754,
+ "▁sustain": 11755,
+ "▁updates": 11756,
+ "▁assign": 11757,
+ "HL": 11758,
+ "▁Sea": 11759,
+ "▁discipl": 11760,
+ "Video": 11761,
+ "▁Chief": 11762,
+ "▁bunch": 11763,
+ "▁Obama": 11764,
+ "nis": 11765,
+ "vor": 11766,
+ "▁agents": 11767,
+ "cas": 11768,
+ "chter": 11769,
+ "▁glanced": 11770,
+ "supported": 11771,
+ "▁Consider": 11772,
+ "▁Everyone": 11773,
+ "▁lect": 11774,
+ "▁Stone": 11775,
+ "▁Jam": 11776,
+ "ogram": 11777,
+ "formance": 11778,
+ "▁\\\"": 11779,
+ "▁patch": 11780,
+ "▁vit": 11781,
+ "Power": 11782,
+ "▁harder": 11783,
+ "Anal": 11784,
+ "▁desired": 11785,
+ "▁jug": 11786,
+ "▁supporting": 11787,
+ "DU": 11788,
+ "]],": 11789,
+ "▁Administr": 11790,
+ "ucky": 11791,
+ "▁controller": 11792,
+ "▁issued": 11793,
+ "▁Sin": 11794,
+ "▁affili": 11795,
+ "▁partners": 11796,
+ "cdots": 11797,
+ "ctic": 11798,
+ "Car": 11799,
+ "▁NY": 11800,
+ "▁priority": 11801,
+ "original": 11802,
+ "Sql": 11803,
+ "▁declared": 11804,
+ "▁Hotel": 11805,
+ "▁browser": 11806,
+ "▁grande": 11807,
+ "}^\\": 11808,
+ "bow": 11809,
+ "▁accommod": 11810,
+ "Directory": 11811,
+ "▁suffering": 11812,
+ "▁logger": 11813,
+ "▁breakfast": 11814,
+ "uli": 11815,
+ "▁boot": 11816,
+ "▁contribution": 11817,
+ "NESS": 11818,
+ "▁Ten": 11819,
+ "semble": 11820,
+ "▁housing": 11821,
+ "Raw": 11822,
+ "ANCE": 11823,
+ "▁При": 11824,
+ "▁brit": 11825,
+ "essa": 11826,
+ "inson": 11827,
+ "▁Ball": 11828,
+ "entes": 11829,
+ "▁Bra": 11830,
+ "score": 11831,
+ "GER": 11832,
+ "route": 11833,
+ "apsed": 11834,
+ "рой": 11835,
+ "diff": 11836,
+ "▁broadcast": 11837,
+ "▁tar": 11838,
+ "▁delight": 11839,
+ ")?": 11840,
+ "chester": 11841,
+ "Platform": 11842,
+ "▁emergency": 11843,
+ "▁ces": 11844,
+ "nership": 11845,
+ "▁situations": 11846,
+ "▁familjen": 11847,
+ "▁Geb": 11848,
+ "enta": 11849,
+ "úblic": 11850,
+ "▁Place": 11851,
+ "ILL": 11852,
+ "▁march": 11853,
+ "▁fundamental": 11854,
+ "attributes": 11855,
+ "кти": 11856,
+ "▁Fu": 11857,
+ "FD": 11858,
+ "▁рас": 11859,
+ "▁academic": 11860,
+ "pres": 11861,
+ "▁rising": 11862,
+ "▁Braz": 11863,
+ "▁receiving": 11864,
+ "WARN": 11865,
+ "▁judg": 11866,
+ "▁necessarily": 11867,
+ "]=": 11868,
+ "▁deeply": 11869,
+ "▁gray": 11870,
+ "Headers": 11871,
+ "▁coal": 11872,
+ "\\{": 11873,
+ "Mut": 11874,
+ "bach": 11875,
+ "▁profit": 11876,
+ "вого": 11877,
+ "igs": 11878,
+ "ograp": 11879,
+ "\";\r": 11880,
+ "▁advoc": 11881,
+ "Generated": 11882,
+ "мери": 11883,
+ "▁Cond": 11884,
+ "▁agric": 11885,
+ "BASE": 11886,
+ "▁arrang": 11887,
+ "▁flowers": 11888,
+ "iw": 11889,
+ "▁];": 11890,
+ "▁вой": 11891,
+ "umerate": 11892,
+ "▁ihr": 11893,
+ "▁пар": 11894,
+ "▁mont": 11895,
+ "widehat": 11896,
+ "mg": 11897,
+ "▁btn": 11898,
+ "▁besk": 11899,
+ "▁acts": 11900,
+ "ós": 11901,
+ "~~~~": 11902,
+ "▁curve": 11903,
+ "language": 11904,
+ "▁TRUE": 11905,
+ "▁cleaning": 11906,
+ "Math": 11907,
+ "▁regional": 11908,
+ "▁estimated": 11909,
+ "arity": 11910,
+ "ierung": 11911,
+ "/{": 11912,
+ "jango": 11913,
+ "$_": 11914,
+ "▁threw": 11915,
+ "rq": 11916,
+ "cop": 11917,
+ "nergy": 11918,
+ "▁Account": 11919,
+ "pal": 11920,
+ "▁Nic": 11921,
+ "]))": 11922,
+ "▁awesome": 11923,
+ "▁Load": 11924,
+ "unnel": 11925,
+ "▁rows": 11926,
+ "▁foreach": 11927,
+ "▁Pod": 11928,
+ "▁EN": 11929,
+ "▁.=": 11930,
+ "uate": 11931,
+ "frastructure": 11932,
+ "▁Watch": 11933,
+ "Stand": 11934,
+ "▁routine": 11935,
+ "▁pic": 11936,
+ "helper": 11937,
+ "▁horses": 11938,
+ "▁requested": 11939,
+ "▁---": 11940,
+ "border": 11941,
+ "▁lifted": 11942,
+ "▁Ped": 11943,
+ "Import": 11944,
+ "ље": 11945,
+ "▁Ли": 11946,
+ "▁myst": 11947,
+ "THER": 11948,
+ "▁AC": 11949,
+ "Proxy": 11950,
+ "prov": 11951,
+ "▁Nik": 11952,
+ "hemat": 11953,
+ "ональ": 11954,
+ "▁\".": 11955,
+ "ului": 11956,
+ "▁improved": 11957,
+ "ieren": 11958,
+ "ocolate": 11959,
+ "Sche": 11960,
+ "unic": 11961,
+ "▁Professor": 11962,
+ "ieler": 11963,
+ "▁duration": 11964,
+ "▁timeout": 11965,
+ "hom": 11966,
+ "▁lux": 11967,
+ "▁trab": 11968,
+ "itary": 11969,
+ "ње": 11970,
+ "▁inspired": 11971,
+ "})\\": 11972,
+ "isely": 11973,
+ "ials": 11974,
+ "▁Vor": 11975,
+ "▁enhance": 11976,
+ "▁lucky": 11977,
+ "World": 11978,
+ "elo": 11979,
+ "ifiers": 11980,
+ "▁facing": 11981,
+ "▁appreciate": 11982,
+ "▁être": 11983,
+ "▁bench": 11984,
+ "atted": 11985,
+ "gence": 11986,
+ "course": 11987,
+ "▁tub": 11988,
+ "▁lors": 11989,
+ "▁mistake": 11990,
+ "nom": 11991,
+ "▁paus": 11992,
+ "▁\"\";": 11993,
+ "▁subs": 11994,
+ "▁stato": 11995,
+ "$)": 11996,
+ "▁gay": 11997,
+ "orry": 11998,
+ "▁vehicles": 11999,
+ "▁brill": 12000,
+ "may": 12001,
+ "resp": 12002,
+ "▁wore": 12003,
+ "ją": 12004,
+ "bp": 12005,
+ "onel": 12006,
+ "▁CR": 12007,
+ "▁diagn": 12008,
+ "mathsf": 12009,
+ "▁holiday": 12010,
+ "▁achieved": 12011,
+ "▁{'": 12012,
+ "▁Resource": 12013,
+ "▁hi": 12014,
+ "▁bra": 12015,
+ "▁CONDITION": 12016,
+ "ctr": 12017,
+ "▁Write": 12018,
+ "ishop": 12019,
+ "OLD": 12020,
+ "▁cpu": 12021,
+ "▁occurs": 12022,
+ "ół": 12023,
+ "straint": 12024,
+ "▁nuclear": 12025,
+ "Area": 12026,
+ "cluster": 12027,
+ "▁surrounding": 12028,
+ "▁Juan": 12029,
+ "▁prima": 12030,
+ "▁Southern": 12031,
+ "itty": 12032,
+ "▁Assembly": 12033,
+ "elem": 12034,
+ "adi": 12035,
+ "éral": 12036,
+ "▁Wat": 12037,
+ "▁Radio": 12038,
+ "▁gegen": 12039,
+ "▁Tony": 12040,
+ "pressed": 12041,
+ "▁Anne": 12042,
+ "▁NS": 12043,
+ "▁Pak": 12044,
+ "▁Civil": 12045,
+ "▁thrown": 12046,
+ "NONE": 12047,
+ "▁pump": 12048,
+ "▁solve": 12049,
+ "ENABLE": 12050,
+ "▁Phys": 12051,
+ "▁],": 12052,
+ "POSE": 12053,
+ "ktet": 12054,
+ "▁Fab": 12055,
+ "validate": 12056,
+ "Iterator": 12057,
+ "condition": 12058,
+ "redu": 12059,
+ "▁negoti": 12060,
+ "anno": 12061,
+ "▁sans": 12062,
+ "▁Ul": 12063,
+ "CHAR": 12064,
+ "▁edition": 12065,
+ "▁spectrum": 12066,
+ "orie": 12067,
+ "▁execution": 12068,
+ "Please": 12069,
+ "▁BO": 12070,
+ "URN": 12071,
+ "▁cow": 12072,
+ "стан": 12073,
+ "istribution": 12074,
+ "Domain": 12075,
+ "▁readers": 12076,
+ "▁consumer": 12077,
+ "▁styles": 12078,
+ "encode": 12079,
+ "▁Cy": 12080,
+ "Common": 12081,
+ "▁Prop": 12082,
+ "▁execute": 12083,
+ "▁eq": 12084,
+ "▁visitors": 12085,
+ "▁Amb": 12086,
+ "udad": 12087,
+ "qquad": 12088,
+ "▁Cert": 12089,
+ "▁trop": 12090,
+ "▁yesterday": 12091,
+ "tain": 12092,
+ "LD": 12093,
+ "atro": 12094,
+ "▁increases": 12095,
+ "▁Wars": 12096,
+ "ned": 12097,
+ "before": 12098,
+ "aupt": 12099,
+ "▁ERR": 12100,
+ "▁Ford": 12101,
+ "▁dalla": 12102,
+ "ULAR": 12103,
+ "▁strike": 12104,
+ "Arr": 12105,
+ "▁recovery": 12106,
+ "▁Response": 12107,
+ "▁strategies": 12108,
+ "▁ін": 12109,
+ "▁rear": 12110,
+ "▁adults": 12111,
+ "▁Не": 12112,
+ "windows": 12113,
+ "decl": 12114,
+ "olen": 12115,
+ "▁Jord": 12116,
+ "▁Kal": 12117,
+ "▁cui": 12118,
+ "▁Про": 12119,
+ "▁Sever": 12120,
+ "▁ale": 12121,
+ "▁peut": 12122,
+ "Stats": 12123,
+ "▁Ross": 12124,
+ "arten": 12125,
+ "shall": 12126,
+ "▁entertain": 12127,
+ "▁parking": 12128,
+ "нови": 12129,
+ "erre": 12130,
+ "▁funding": 12131,
+ "▁Cle": 12132,
+ "▁Ot": 12133,
+ "unst": 12134,
+ "assertEquals": 12135,
+ "▁cancell": 12136,
+ "TAG": 12137,
+ "▁Early": 12138,
+ "▁feedback": 12139,
+ "▁pand": 12140,
+ "yo": 12141,
+ "▁mirror": 12142,
+ "▁verb": 12143,
+ "▁highlight": 12144,
+ "erialize": 12145,
+ "▁grade": 12146,
+ "лась": 12147,
+ "▁Brook": 12148,
+ "▁LI": 12149,
+ "▁implies": 12150,
+ "▁enorm": 12151,
+ "ają": 12152,
+ "▁Wer": 12153,
+ "away": 12154,
+ "▁machines": 12155,
+ "▁dent": 12156,
+ "Idx": 12157,
+ "▁tid": 12158,
+ ")\"": 12159,
+ "▁mole": 12160,
+ "bold": 12161,
+ "CONT": 12162,
+ "▁ép": 12163,
+ "▁cutting": 12164,
+ "▁Neg": 12165,
+ "▁tong": 12166,
+ "▁networks": 12167,
+ "▁Fall": 12168,
+ "generated": 12169,
+ "▁Pri": 12170,
+ "UEST": 12171,
+ "▁Belg": 12172,
+ "▁sheet": 12173,
+ "кси": 12174,
+ "▁†": 12175,
+ "▁yeah": 12176,
+ "▁Victor": 12177,
+ "▁Rub": 12178,
+ "▁candidates": 12179,
+ "prés": 12180,
+ "▁EU": 12181,
+ "etr": 12182,
+ "▁rolled": 12183,
+ "▁Pas": 12184,
+ "▁Arthur": 12185,
+ "Arch": 12186,
+ "▁Mann": 12187,
+ "American": 12188,
+ "zes": 12189,
+ "inners": 12190,
+ "▁Auto": 12191,
+ "▁professor": 12192,
+ "▁);\r": 12193,
+ "▁addr": 12194,
+ "▁Medical": 12195,
+ "▁fired": 12196,
+ "▁Core": 12197,
+ "▁CONFIG": 12198,
+ "▁sql": 12199,
+ "▁Conserv": 12200,
+ "ichen": 12201,
+ "Vertex": 12202,
+ "▁HO": 12203,
+ "Yeah": 12204,
+ "Note": 12205,
+ "▁OK": 12206,
+ "mus": 12207,
+ "focus": 12208,
+ "aja": 12209,
+ "rá": 12210,
+ "▁hence": 12211,
+ "▁executive": 12212,
+ "▁liquid": 12213,
+ "uje": 12214,
+ "▁driven": 12215,
+ "igue": 12216,
+ "▁Wik": 12217,
+ "Rate": 12218,
+ "rand": 12219,
+ "Results": 12220,
+ "▁copies": 12221,
+ "▁tan": 12222,
+ "riteria": 12223,
+ "enen": 12224,
+ "}_\\": 12225,
+ "▁pobl": 12226,
+ "▁southern": 12227,
+ "eln": 12228,
+ "▁zwei": 12229,
+ "▁concrete": 12230,
+ "▁CONDITIONS": 12231,
+ "▁dreams": 12232,
+ "▁minim": 12233,
+ "▁employee": 12234,
+ "▁nap": 12235,
+ "▁suspect": 12236,
+ "Mouse": 12237,
+ "▁therapy": 12238,
+ "aval": 12239,
+ "▁Anth": 12240,
+ "START": 12241,
+ "sters": 12242,
+ "ishment": 12243,
+ "finite": 12244,
+ "WA": 12245,
+ "vy": 12246,
+ "▁mood": 12247,
+ "comfort": 12248,
+ "▁shr": 12249,
+ "▁decade": 12250,
+ "ября": 12251,
+ "▁'#": 12252,
+ "▁dot": 12253,
+ "▁hill": 12254,
+ "arry": 12255,
+ "catch": 12256,
+ "▁jQuery": 12257,
+ "▁corporate": 12258,
+ "▁BASIS": 12259,
+ "▁appointed": 12260,
+ "▁embar": 12261,
+ "ographie": 12262,
+ "▁pressed": 12263,
+ "▁champion": 12264,
+ "emit": 12265,
+ "▁Bed": 12266,
+ "вання": 12267,
+ "Gui": 12268,
+ "▁PUR": 12269,
+ "▁urban": 12270,
+ "▁sentence": 12271,
+ "bury": 12272,
+ "▁Video": 12273,
+ "▁regularly": 12274,
+ "vl": 12275,
+ "▁слу": 12276,
+ "ockey": 12277,
+ "evin": 12278,
+ "ultural": 12279,
+ "▁passage": 12280,
+ "▁состав": 12281,
+ "▁largely": 12282,
+ "orters": 12283,
+ "▁connections": 12284,
+ "▁surprising": 12285,
+ "bc": 12286,
+ "▁strongly": 12287,
+ "ansas": 12288,
+ "▁sist": 12289,
+ "▁extreme": 12290,
+ "whel": 12291,
+ "▁dealing": 12292,
+ "ographic": 12293,
+ "▁Republican": 12294,
+ "▁granted": 12295,
+ "▁CL": 12296,
+ "▁Hope": 12297,
+ "lessly": 12298,
+ "▁upload": 12299,
+ "▁-\\": 12300,
+ "нию": 12301,
+ "▁valuable": 12302,
+ "=[": 12303,
+ "Price": 12304,
+ "issance": 12305,
+ "iens": 12306,
+ "heit": 12307,
+ "▁suggests": 12308,
+ "сло": 12309,
+ "▁jur": 12310,
+ "}|": 12311,
+ "lp": 12312,
+ "▁invited": 12313,
+ "▁deriv": 12314,
+ "IMIT": 12315,
+ "rass": 12316,
+ "▁instruct": 12317,
+ "▁courses": 12318,
+ "äch": 12319,
+ "▁fifty": 12320,
+ "DEVICE": 12321,
+ "ASH": 12322,
+ "▁hip": 12323,
+ "Unknown": 12324,
+ "▁Catalogue": 12325,
+ "▁Roll": 12326,
+ "▁tensor": 12327,
+ "bec": 12328,
+ "été": 12329,
+ "Identity": 12330,
+ "&\\": 12331,
+ "▁Stephen": 12332,
+ "nodes": 12333,
+ "Dim": 12334,
+ "▁consists": 12335,
+ "▁normally": 12336,
+ "ubl": 12337,
+ "▁Police": 12338,
+ "▁Games": 12339,
+ "five": 12340,
+ "Have": 12341,
+ "▁padding": 12342,
+ "eres": 12343,
+ "anth": 12344,
+ "▁puts": 12345,
+ "uminate": 12346,
+ "ovie": 12347,
+ "▁Index": 12348,
+ "blue": 12349,
+ "Scal": 12350,
+ "▁giant": 12351,
+ "TF": 12352,
+ "pson": 12353,
+ "▁victim": 12354,
+ "serial": 12355,
+ "▁Sym": 12356,
+ "Single": 12357,
+ "▁md": 12358,
+ "▁attended": 12359,
+ "▁Stra": 12360,
+ "▁Dark": 12361,
+ ")|": 12362,
+ "▁span": 12363,
+ "▁maintenance": 12364,
+ "▁bind": 12365,
+ "Bean": 12366,
+ "ilarly": 12367,
+ "▁convent": 12368,
+ "▁José": 12369,
+ "udd": 12370,
+ "▁poly": 12371,
+ "▁idx": 12372,
+ "▁asks": 12373,
+ "▁enthus": 12374,
+ "▁suck": 12375,
+ "▁Cou": 12376,
+ "▁Corporation": 12377,
+ "usions": 12378,
+ "opher": 12379,
+ "▁symptoms": 12380,
+ "▁Johann": 12381,
+ "▁пу": 12382,
+ "▁html": 12383,
+ "▁ps": 12384,
+ "earing": 12385,
+ "gesch": 12386,
+ "▁Mother": 12387,
+ "RET": 12388,
+ "▁furniture": 12389,
+ "PF": 12390,
+ "▁Guard": 12391,
+ "pattern": 12392,
+ "▁lovely": 12393,
+ "alg": 12394,
+ "edly": 12395,
+ "sex": 12396,
+ "▁finds": 12397,
+ "Buf": 12398,
+ "▁над": 12399,
+ "▁км": 12400,
+ "▁Por": 12401,
+ "СР": 12402,
+ "Enter": 12403,
+ "▁esta": 12404,
+ "▁тре": 12405,
+ "▁\"*": 12406,
+ "▁Fox": 12407,
+ "▁cock": 12408,
+ "Bundle": 12409,
+ "▁puis": 12410,
+ "▁announce": 12411,
+ "▁guid": 12412,
+ "checked": 12413,
+ "icide": 12414,
+ "neg": 12415,
+ "▁Gil": 12416,
+ "schen": 12417,
+ "ologist": 12418,
+ "iso": 12419,
+ "groups": 12420,
+ "▁somebody": 12421,
+ "Day": 12422,
+ "tras": 12423,
+ "▁compact": 12424,
+ "▁organized": 12425,
+ "▁roles": 12426,
+ "▁hint": 12427,
+ "▁så": 12428,
+ "▁pays": 12429,
+ "▁Си": 12430,
+ "▁hoped": 12431,
+ "▁sail": 12432,
+ "▁Vers": 12433,
+ "▁embr": 12434,
+ "▁bot": 12435,
+ "▁exceed": 12436,
+ "BACK": 12437,
+ "▁gaze": 12438,
+ "▁spons": 12439,
+ "AST": 12440,
+ "▁torch": 12441,
+ "▁newspaper": 12442,
+ "▁Dist": 12443,
+ "▁bass": 12444,
+ "▁hanging": 12445,
+ "▁ears": 12446,
+ "ńsk": 12447,
+ "getValue": 12448,
+ "▁unus": 12449,
+ "▁Ele": 12450,
+ "services": 12451,
+ "▁dressed": 12452,
+ "lav": 12453,
+ "▁пла": 12454,
+ "Private": 12455,
+ "mic": 12456,
+ "▁parser": 12457,
+ "▁sections": 12458,
+ "▁fo": 12459,
+ "Errorf": 12460,
+ "inz": 12461,
+ "örd": 12462,
+ "▁metric": 12463,
+ "URI": 12464,
+ "▁vice": 12465,
+ "RED": 12466,
+ "▁nue": 12467,
+ "revs": 12468,
+ "▁collected": 12469,
+ "oose": 12470,
+ "▁mond": 12471,
+ "▁nas": 12472,
+ "▁Насе": 12473,
+ "▁å": 12474,
+ "Drop": 12475,
+ "▁abuse": 12476,
+ "▁sees": 12477,
+ "▁Hence": 12478,
+ "exec": 12479,
+ "}\\,": 12480,
+ "▁arbitr": 12481,
+ "▁Application": 12482,
+ "family": 12483,
+ "üd": 12484,
+ "▁magnetic": 12485,
+ "▁newly": 12486,
+ "▁reprodu": 12487,
+ "▁writers": 12488,
+ "▁headers": 12489,
+ "ší": 12490,
+ "рт": 12491,
+ "YPE": 12492,
+ "▁schema": 12493,
+ "▁Ce": 12494,
+ "▁Jews": 12495,
+ "▁Record": 12496,
+ "present": 12497,
+ "▁также": 12498,
+ "▁labels": 12499,
+ "Socket": 12500,
+ "▁equations": 12501,
+ "▁medicine": 12502,
+ "▁authorities": 12503,
+ "}`": 12504,
+ "стви": 12505,
+ "▁Corn": 12506,
+ "▁environmental": 12507,
+ "WARE": 12508,
+ "Mer": 12509,
+ "▁само": 12510,
+ "▁Technology": 12511,
+ "▁Saf": 12512,
+ "▁conn": 12513,
+ "▁Um": 12514,
+ "▁Pacific": 12515,
+ "тел": 12516,
+ "jan": 12517,
+ "▁uncertain": 12518,
+ "▁belief": 12519,
+ "counter": 12520,
+ "toBe": 12521,
+ "INS": 12522,
+ "weet": 12523,
+ "Light": 12524,
+ "primary": 12525,
+ "▁featured": 12526,
+ "▁touched": 12527,
+ "HTTP": 12528,
+ "▁tact": 12529,
+ "pository": 12530,
+ "▁eines": 12531,
+ "lass": 12532,
+ "ська": 12533,
+ "▁przez": 12534,
+ "▁fuer": 12535,
+ "▁exciting": 12536,
+ "▁Cub": 12537,
+ "agan": 12538,
+ "VO": 12539,
+ "▁'%": 12540,
+ "▁\\{": 12541,
+ "ubble": 12542,
+ "▁Fol": 12543,
+ "▁Kong": 12544,
+ "▁versch": 12545,
+ "FAIL": 12546,
+ "▁naar": 12547,
+ "ös": 12548,
+ "speed": 12549,
+ "▁territor": 12550,
+ "▁wrap": 12551,
+ "▁Jahre": 12552,
+ "lee": 12553,
+ "▁crossed": 12554,
+ "resolve": 12555,
+ "▁stim": 12556,
+ "Native": 12557,
+ "ursor": 12558,
+ "NotNull": 12559,
+ "▁Albert": 12560,
+ "▁signature": 12561,
+ "▁Ru": 12562,
+ "idas": 12563,
+ "▁decent": 12564,
+ "▁faced": 12565,
+ "▁лю": 12566,
+ "▁Spain": 12567,
+ "▁resistance": 12568,
+ "▁Brian": 12569,
+ "kwargs": 12570,
+ "▁interval": 12571,
+ "▁Ле": 12572,
+ "▁explo": 12573,
+ "▁semi": 12574,
+ "▁widely": 12575,
+ "dx": 12576,
+ "kov": 12577,
+ "▁Come": 12578,
+ "▁knife": 12579,
+ "Asp": 12580,
+ "uno": 12581,
+ "lineto": 12582,
+ "▁Bund": 12583,
+ "Cert": 12584,
+ "▁todo": 12585,
+ "tags": 12586,
+ "▁guarantee": 12587,
+ "▁vital": 12588,
+ "▁fought": 12589,
+ "▁Env": 12590,
+ "HD": 12591,
+ "Lower": 12592,
+ "Tx": 12593,
+ "▁Fa": 12594,
+ "▁anticip": 12595,
+ "Timer": 12596,
+ "mediate": 12597,
+ "▁proven": 12598,
+ "▁partir": 12599,
+ "AE": 12600,
+ "cursor": 12601,
+ "▁wooden": 12602,
+ "▁Contact": 12603,
+ "regs": 12604,
+ "▁provinc": 12605,
+ "▁DC": 12606,
+ "▁memories": 12607,
+ "▁ft": 12608,
+ "▁battery": 12609,
+ "utenant": 12610,
+ "Login": 12611,
+ "ountry": 12612,
+ "▁compens": 12613,
+ "operatorname": 12614,
+ "▁Jacob": 12615,
+ "zed": 12616,
+ "ADDR": 12617,
+ "▁quad": 12618,
+ "*).": 12619,
+ "▁coat": 12620,
+ "▁fir": 12621,
+ "▁Michel": 12622,
+ "▁Standard": 12623,
+ "rf": 12624,
+ "mel": 12625,
+ "▁coeff": 12626,
+ "▁Iraq": 12627,
+ "▁Given": 12628,
+ "нима": 12629,
+ "▁FIT": 12630,
+ "▁peu": 12631,
+ "▁ig": 12632,
+ "▁Case": 12633,
+ "mé": 12634,
+ "▁parallel": 12635,
+ "cio": 12636,
+ "kow": 12637,
+ "▁institutions": 12638,
+ "ícul": 12639,
+ "aban": 12640,
+ "UX": 12641,
+ "▁Sarah": 12642,
+ "▁més": 12643,
+ "▁atmos": 12644,
+ "▁släktet": 12645,
+ "▁brothers": 12646,
+ "▁wanting": 12647,
+ "aaaa": 12648,
+ "▁fest": 12649,
+ "=-": 12650,
+ "▁forty": 12651,
+ "▁creates": 12652,
+ "hh": 12653,
+ "▁Android": 12654,
+ "anches": 12655,
+ "BT": 12656,
+ "upload": 12657,
+ "xis": 12658,
+ "Hz": 12659,
+ "бор": 12660,
+ "RAY": 12661,
+ "ntil": 12662,
+ "▁leaned": 12663,
+ "unda": 12664,
+ "▁ultimately": 12665,
+ "▁tok": 12666,
+ "neh": 12667,
+ "▁lawyer": 12668,
+ "hend": 12669,
+ "▁Vin": 12670,
+ "▁facility": 12671,
+ "▁likes": 12672,
+ "ento": 12673,
+ "Nodes": 12674,
+ "▁entrance": 12675,
+ "atto": 12676,
+ "rett": 12677,
+ "accept": 12678,
+ "theme": 12679,
+ "тан": 12680,
+ "osi": 12681,
+ "▁{},": 12682,
+ "pgfpathlineto": 12683,
+ "good": 12684,
+ "slot": 12685,
+ "▁innoc": 12686,
+ "▁proport": 12687,
+ "▁arrive": 12688,
+ "ého": 12689,
+ "▁pairs": 12690,
+ "▁wrapped": 12691,
+ "▁unw": 12692,
+ "▁explos": 12693,
+ "▁gel": 12694,
+ "Will": 12695,
+ "▁Zealand": 12696,
+ "ías": 12697,
+ "▁Jr": 12698,
+ "▁Fra": 12699,
+ "▁legit": 12700,
+ "▁illegal": 12701,
+ "клю": 12702,
+ "▁tort": 12703,
+ "▁pron": 12704,
+ "Fi": 12705,
+ "▁forg": 12706,
+ "export": 12707,
+ "▁Children": 12708,
+ "▁Abs": 12709,
+ "▁Send": 12710,
+ "▁discount": 12711,
+ "▁poster": 12712,
+ "ented": 12713,
+ "anim": 12714,
+ "verb": 12715,
+ "sto": 12716,
+ "▁Bible": 12717,
+ "pending": 12718,
+ "▁Phot": 12719,
+ "strap": 12720,
+ "ieron": 12721,
+ "PG": 12722,
+ "cular": 12723,
+ "crit": 12724,
+ "urd": 12725,
+ "ENO": 12726,
+ "▁northern": 12727,
+ "▁naturally": 12728,
+ "<'": 12729,
+ "weg": 12730,
+ "▁drunk": 12731,
+ "▁Dal": 12732,
+ "▁mouse": 12733,
+ "▁continuous": 12734,
+ "▁initially": 12735,
+ "agu": 12736,
+ "мпи": 12737,
+ "ANT": 12738,
+ "Div": 12739,
+ "▁recording": 12740,
+ "Bind": 12741,
+ "▁correctly": 12742,
+ "initial": 12743,
+ "▁Rights": 12744,
+ "▁debate": 12745,
+ "WRITE": 12746,
+ "built": 12747,
+ "▁permit": 12748,
+ "▁professionals": 12749,
+ "cv": 12750,
+ "▁DI": 12751,
+ "▁handed": 12752,
+ "▁Cu": 12753,
+ "▁Hospital": 12754,
+ "▁beskrevs": 12755,
+ "ней": 12756,
+ "ност": 12757,
+ "▁anxiety": 12758,
+ "▁heavily": 12759,
+ "▁Var": 12760,
+ "▁dispos": 12761,
+ "+\"": 12762,
+ "▁Ever": 12763,
+ "izon": 12764,
+ "▁operators": 12765,
+ "nego": 12766,
+ "▁Bry": 12767,
+ "▁votes": 12768,
+ "izione": 12769,
+ "▁рай": 12770,
+ "▁feat": 12771,
+ "▁western": 12772,
+ "▁confront": 12773,
+ "▁stronger": 12774,
+ "▁фа": 12775,
+ "stre": 12776,
+ "▁Valid": 12777,
+ "▁nad": 12778,
+ "▁checking": 12779,
+ "▁birds": 12780,
+ "▁Northern": 12781,
+ "▁intention": 12782,
+ "uce": 12783,
+ "▁covers": 12784,
+ "▁wondering": 12785,
+ "▁Optional": 12786,
+ "protocol": 12787,
+ "▁aggress": 12788,
+ "——": 12789,
+ "Vec": 12790,
+ "▁dates": 12791,
+ "quot": 12792,
+ "▁bom": 12793,
+ "▁scan": 12794,
+ "▁Item": 12795,
+ "▁Navy": 12796,
+ "▁Gran": 12797,
+ "▁everybody": 12798,
+ "▁unexpected": 12799,
+ "▁divor": 12800,
+ "▁ease": 12801,
+ "umbled": 12802,
+ "^+": 12803,
+ "cuss": 12804,
+ "▁pale": 12805,
+ "▁Inga": 12806,
+ "▁Broad": 12807,
+ "▁Medic": 12808,
+ "▁Roy": 12809,
+ "▁Inn": 12810,
+ "▁pens": 12811,
+ "PN": 12812,
+ ".:": 12813,
+ "▁principle": 12814,
+ "▁letting": 12815,
+ "▁conducted": 12816,
+ "FALSE": 12817,
+ "▁OS": 12818,
+ "Focus": 12819,
+ "▁measured": 12820,
+ "▁Democratic": 12821,
+ "High": 12822,
+ "▁pré": 12823,
+ "ennes": 12824,
+ "▁indicates": 12825,
+ "▁ending": 12826,
+ "▁Small": 12827,
+ "▁": 26345,
+ "olent": 26346,
+ "▁этого": 26347,
+ "▁Generic": 26348,
+ "▁*/,": 26349,
+ "▁combinations": 26350,
+ "▁rejo": 26351,
+ "спубли": 26352,
+ "capacity": 26353,
+ "▁traces": 26354,
+ "▁opacity": 26355,
+ "▁Official": 26356,
+ "icion": 26357,
+ "▁emotionally": 26358,
+ "▁Joel": 26359,
+ "ському": 26360,
+ "▁legendary": 26361,
+ "▁pam": 26362,
+ "▁También": 26363,
+ ".<": 26364,
+ "iba": 26365,
+ "midt": 26366,
+ "бом": 26367,
+ "▁ensuite": 26368,
+ "Authorization": 26369,
+ "Pag": 26370,
+ "▁helmet": 26371,
+ "▁territo": 26372,
+ "secondary": 26373,
+ "▁segunda": 26374,
+ "▁Wire": 26375,
+ "recated": 26376,
+ "▁invoked": 26377,
+ "▁ValueError": 26378,
+ "▁фо": 26379,
+ "ALIGN": 26380,
+ "CURRENT": 26381,
+ "\\+\\_\\": 26382,
+ "▁compilation": 26383,
+ "ær": 26384,
+ "▁Palmar": 26385,
+ "▁influences": 26386,
+ "/:": 26387,
+ "Mix": 26388,
+ "NOP": 26389,
+ "econom": 26390,
+ "▁tucked": 26391,
+ "▁});\r": 26392,
+ "ANK": 26393,
+ "reject": 26394,
+ "▁pension": 26395,
+ "▁generates": 26396,
+ "чё": 26397,
+ "▁incap": 26398,
+ "▁clicked": 26399,
+ "▁fus": 26400,
+ "ourses": 26401,
+ "▁Easter": 26402,
+ "%;": 26403,
+ "zin": 26404,
+ "▁obligations": 26405,
+ "▁Tips": 26406,
+ "};\r": 26407,
+ ".\"_": 26408,
+ "▁BSD": 26409,
+ "ática": 26410,
+ "▁expose": 26411,
+ "Pars": 26412,
+ "▁Amanda": 26413,
+ "куп": 26414,
+ "▁guessed": 26415,
+ "dsi": 26416,
+ "▁Leip": 26417,
+ "Broad": 26418,
+ "▁Hughes": 26419,
+ "ié": 26420,
+ "▁Wahl": 26421,
+ "▁formerly": 26422,
+ "Relative": 26423,
+ "▁Yu": 26424,
+ "▁Mountains": 26425,
+ "▁Enum": 26426,
+ "▁strang": 26427,
+ "_-": 26428,
+ "recht": 26429,
+ "viv": 26430,
+ "pause": 26431,
+ "▁Londres": 26432,
+ "▁elbow": 26433,
+ "▁Hawaii": 26434,
+ "▁Casino": 26435,
+ "Threshold": 26436,
+ "Units": 26437,
+ "Include": 26438,
+ "ито": 26439,
+ "asury": 26440,
+ "▁steht": 26441,
+ "▁damned": 26442,
+ "▁packets": 26443,
+ "▁Werk": 26444,
+ "▁elevator": 26445,
+ "iedad": 26446,
+ "govern": 26447,
+ "▁CONTRACT": 26448,
+ "mals": 26449,
+ "▁remem": 26450,
+ "▁entonces": 26451,
+ "▁vas": 26452,
+ "▁sympathy": 26453,
+ "▁befindet": 26454,
+ "incing": 26455,
+ "DataSet": 26456,
+ "▁additionally": 26457,
+ "▁musician": 26458,
+ "шего": 26459,
+ "▁listop": 26460,
+ ">\")": 26461,
+ "Printf": 26462,
+ "▁Felix": 26463,
+ "▁carved": 26464,
+ "▁nicely": 26465,
+ "гом": 26466,
+ "chap": 26467,
+ "▁Nieder": 26468,
+ "▁Lav": 26469,
+ "▁modifications": 26470,
+ "moment": 26471,
+ "▁balcon": 26472,
+ "▁dependency": 26473,
+ "CKET": 26474,
+ "▁vanished": 26475,
+ "▁fighters": 26476,
+ "▁zunächst": 26477,
+ "ioctl": 26478,
+ "▁defens": 26479,
+ "▁Nem": 26480,
+ "Utility": 26481,
+ "▁curv": 26482,
+ "▁DAMAGES": 26483,
+ "▁Rogers": 26484,
+ "▁gratitude": 26485,
+ "▁Denmark": 26486,
+ "рая": 26487,
+ "grpc": 26488,
+ "▁juni": 26489,
+ "▁октября": 26490,
+ "▁immense": 26491,
+ "▁prevented": 26492,
+ "▁foam": 26493,
+ "▁Extra": 26494,
+ "aimed": 26495,
+ "▁Criteria": 26496,
+ "▁Simply": 26497,
+ "boxes": 26498,
+ "▁Legend": 26499,
+ "▁Players": 26500,
+ "▁Mercedes": 26501,
+ "▁Branch": 26502,
+ "TERN": 26503,
+ "omena": 26504,
+ "▁incorporate": 26505,
+ "conde": 26506,
+ "▁Estado": 26507,
+ "▁wasted": 26508,
+ "▁complaining": 26509,
+ "▁warriors": 26510,
+ "oter": 26511,
+ "▁этом": 26512,
+ "▁conten": 26513,
+ "▁machinery": 26514,
+ "▁technological": 26515,
+ "▁TD": 26516,
+ "▁gras": 26517,
+ "▁minimize": 26518,
+ "▁Door": 26519,
+ "▁bzw": 26520,
+ "▁prac": 26521,
+ "TREE": 26522,
+ "▁Wing": 26523,
+ "▁Transaction": 26524,
+ "▁MVT": 26525,
+ "▁Klein": 26526,
+ "commons": 26527,
+ "▁}{": 26528,
+ "▁Heritage": 26529,
+ "▁fade": 26530,
+ "рок": 26531,
+ "setValue": 26532,
+ "▁Wallace": 26533,
+ "MX": 26534,
+ "▁ACT": 26535,
+ "▁footage": 26536,
+ "▁entstand": 26537,
+ "arga": 26538,
+ "▁nails": 26539,
+ "▁capitalism": 26540,
+ "▁Garc": 26541,
+ "▁suspension": 26542,
+ "ilis": 26543,
+ "▁Mov": 26544,
+ "uffled": 26545,
+ "Arc": 26546,
+ "▁Beautiful": 26547,
+ "WAY": 26548,
+ "Parallel": 26549,
+ "XXXX": 26550,
+ "diag": 26551,
+ "▁DT": 26552,
+ "mq": 26553,
+ "TextView": 26554,
+ "MLE": 26555,
+ "ennen": 26556,
+ "▁infected": 26557,
+ "▁therapist": 26558,
+ "INGS": 26559,
+ "▁cidade": 26560,
+ "ън": 26561,
+ "▁pdf": 26562,
+ "▁bump": 26563,
+ "CTX": 26564,
+ "▁INCLUDING": 26565,
+ "▁Gef": 26566,
+ "ENTIAL": 26567,
+ "▁handy": 26568,
+ "▁temporal": 26569,
+ "AtA": 26570,
+ "ISH": 26571,
+ "▁Pattern": 26572,
+ "▁lan": 26573,
+ "ependant": 26574,
+ "▁shining": 26575,
+ "idy": 26576,
+ "▁NT": 26577,
+ "▁Fran": 26578,
+ "▁nurses": 26579,
+ "▁betray": 26580,
+ "▁sensible": 26581,
+ "▁апреля": 26582,
+ "▁'[": 26583,
+ "▁thirteen": 26584,
+ ")}_{": 26585,
+ "▁Noah": 26586,
+ "INSERT": 26587,
+ "istically": 26588,
+ "▁Appendix": 26589,
+ "▁recher": 26590,
+ "Receiver": 26591,
+ "▁dernier": 26592,
+ "лла": 26593,
+ "лиза": 26594,
+ "▁Partido": 26595,
+ "▁maximal": 26596,
+ "snap": 26597,
+ "▁часть": 26598,
+ "STOP": 26599,
+ "▁ultra": 26600,
+ "▁développ": 26601,
+ "▁tegen": 26602,
+ "▁Чи": 26603,
+ "LIB": 26604,
+ "▁baseline": 26605,
+ "reload": 26606,
+ "▁Arbitro": 26607,
+ "▁kall": 26608,
+ "capture": 26609,
+ "Arm": 26610,
+ "quin": 26611,
+ "impse": 26612,
+ "zas": 26613,
+ "▁Cand": 26614,
+ "▁brains": 26615,
+ "▁hostile": 26616,
+ "▁marble": 26617,
+ "oons": 26618,
+ "▁Loss": 26619,
+ "MetaData": 26620,
+ "▁República": 26621,
+ "▁andra": 26622,
+ "oden": 26623,
+ "▁documented": 26624,
+ "▁Moses": 26625,
+ "odd": 26626,
+ "▁wax": 26627,
+ "usch": 26628,
+ "▁diagnosed": 26629,
+ "inkle": 26630,
+ "▁Xbox": 26631,
+ "▁seventy": 26632,
+ "cias": 26633,
+ "▁noviembre": 26634,
+ "Compute": 26635,
+ "});\r": 26636,
+ "▁Philippe": 26637,
+ "▁För": 26638,
+ "Leave": 26639,
+ "▁sage": 26640,
+ "▁unpre": 26641,
+ "▁Fortunately": 26642,
+ "▁apost": 26643,
+ "entities": 26644,
+ "▁ellos": 26645,
+ "authorized": 26646,
+ "GBT": 26647,
+ "▁insist": 26648,
+ "▁inspire": 26649,
+ "Mass": 26650,
+ "▁rôle": 26651,
+ "fee": 26652,
+ "ipart": 26653,
+ "цер": 26654,
+ "unate": 26655,
+ "▁CNN": 26656,
+ ":}": 26657,
+ "▁unhappy": 26658,
+ "▁imported": 26659,
+ "HIGH": 26660,
+ "rings": 26661,
+ "▁Instance": 26662,
+ "Bay": 26663,
+ "agles": 26664,
+ "mee": 26665,
+ "bery": 26666,
+ "▁Stories": 26667,
+ "▁Chase": 26668,
+ "▁carriage": 26669,
+ "▁misunder": 26670,
+ "▁imagin": 26671,
+ "pw": 26672,
+ "▁Meter": 26673,
+ "▁crowds": 26674,
+ "▁Fame": 26675,
+ "skill": 26676,
+ "▁comed": 26677,
+ "▁ranch": 26678,
+ "▁lacking": 26679,
+ "▁submar": 26680,
+ "iante": 26681,
+ "▁lanz": 26682,
+ "▁служ": 26683,
+ "-----------": 26684,
+ "▁obten": 26685,
+ "▁downstairs": 26686,
+ "YN": 26687,
+ "rotation": 26688,
+ "▁Jesse": 26689,
+ "$(\"#": 26690,
+ "▁puls": 26691,
+ "irling": 26692,
+ "▁Schaus": 26693,
+ "▁deployed": 26694,
+ "▁{}\",": 26695,
+ "▁Marvel": 26696,
+ "ENUM": 26697,
+ "▁Mathemat": 26698,
+ "▁nn": 26699,
+ "compet": 26700,
+ "ków": 26701,
+ "bil": 26702,
+ "Which": 26703,
+ "isine": 26704,
+ "▁rude": 26705,
+ "▁niveau": 26706,
+ "▁área": 26707,
+ "▁près": 26708,
+ "atis": 26709,
+ "▁[...]": 26710,
+ "fur": 26711,
+ "omm": 26712,
+ "packed": 26713,
+ "мене": 26714,
+ "scriptstyle": 26715,
+ "▁Ath": 26716,
+ "▁desp": 26717,
+ "eltemperaturen": 26718,
+ "▁talents": 26719,
+ "ocy": 26720,
+ "▁raises": 26721,
+ "LIMIT": 26722,
+ "▁editorial": 26723,
+ "▁Animal": 26724,
+ "drive": 26725,
+ "▁работа": 26726,
+ "bss": 26727,
+ "▁Sev": 26728,
+ "epoch": 26729,
+ "▁RC": 26730,
+ "UNUSED": 26731,
+ "▁mandatory": 26732,
+ "(?:": 26733,
+ "▁Bin": 26734,
+ "▁synthetic": 26735,
+ "▁gown": 26736,
+ "▁Dob": 26737,
+ "kap": 26738,
+ "▁harmon": 26739,
+ "▁liberty": 26740,
+ "▁Rice": 26741,
+ "▁prayers": 26742,
+ "▁mise": 26743,
+ "▁confusing": 26744,
+ "▁leap": 26745,
+ "▁arrives": 26746,
+ "kamp": 26747,
+ "▁thats": 26748,
+ "ACC": 26749,
+ "▁Parameters": 26750,
+ "▁одно": 26751,
+ "▁Bio": 26752,
+ "density": 26753,
+ "▁glimpse": 26754,
+ "FORE": 26755,
+ "▁Listen": 26756,
+ "Prev": 26757,
+ "}\\,\\": 26758,
+ "куль": 26759,
+ "▁SEC": 26760,
+ "▁explored": 26761,
+ "▁meantime": 26762,
+ "AIL": 26763,
+ "▁WP": 26764,
+ "▁raison": 26765,
+ "▁existe": 26766,
+ "▁lesser": 26767,
+ "▁Validate": 26768,
+ "▁caution": 26769,
+ "usta": 26770,
+ "heading": 26771,
+ "EFF": 26772,
+ ".'\"": 26773,
+ "▁Gilbert": 26774,
+ "▁limitation": 26775,
+ "▁retour": 26776,
+ "▁Commonwealth": 26777,
+ "▁gewann": 26778,
+ "▁miserable": 26779,
+ "▁networking": 26780,
+ "▁ottobre": 26781,
+ "▁Dise": 26782,
+ "edges": 26783,
+ "▁sede": 26784,
+ "вича": 26785,
+ "uniform": 26786,
+ "▁деятель": 26787,
+ "iros": 26788,
+ "▁desen": 26789,
+ "▁parc": 26790,
+ "▁Rico": 26791,
+ "Ns": 26792,
+ "guid": 26793,
+ "orio": 26794,
+ "avelength": 26795,
+ "▁Gle": 26796,
+ "inceton": 26797,
+ "Amaz": 26798,
+ "Construct": 26799,
+ "▁mx": 26800,
+ "▁Vern": 26801,
+ "▁Generation": 26802,
+ "Jack": 26803,
+ "romag": 26804,
+ "▁viagra": 26805,
+ "▁Peg": 26806,
+ "▁Updated": 26807,
+ "▁overlap": 26808,
+ "EventArgs": 26809,
+ "кро": 26810,
+ "▁*«": 26811,
+ "▁questioned": 26812,
+ "South": 26813,
+ "notice": 26814,
+ "▁permanently": 26815,
+ "lst": 26816,
+ "ficie": 26817,
+ "▁quella": 26818,
+ "▁colleges": 26819,
+ "▁disappointment": 26820,
+ "▁Luft": 26821,
+ "imgur": 26822,
+ "▁transitions": 26823,
+ "▁seller": 26824,
+ "▁июня": 26825,
+ "▁Og": 26826,
+ "▁ADD": 26827,
+ "▁Pays": 26828,
+ "COMMAND": 26829,
+ "grades": 26830,
+ "▁febbra": 26831,
+ "▁Cyr": 26832,
+ "▁febbraio": 26833,
+ "eti": 26834,
+ "▁arom": 26835,
+ "▁Claude": 26836,
+ "▁UEFA": 26837,
+ "▁живе": 26838,
+ "▁Victorian": 26839,
+ "keeping": 26840,
+ "ên": 26841,
+ "▁FIXME": 26842,
+ "itime": 26843,
+ "chestr": 26844,
+ "▁Samsung": 26845,
+ "▁doctrine": 26846,
+ "▁pear": 26847,
+ "▁Mediterranean": 26848,
+ "▁Ya": 26849,
+ "▁vault": 26850,
+ "▁Historic": 26851,
+ "▁sedan": 26852,
+ "▁heated": 26853,
+ "▁política": 26854,
+ "Proof": 26855,
+ ":{": 26856,
+ "fem": 26857,
+ "▁Frankfurt": 26858,
+ "pectives": 26859,
+ "MG": 26860,
+ "▁Eye": 26861,
+ "dai": 26862,
+ "▁reserves": 26863,
+ "NER": 26864,
+ "▁tobacco": 26865,
+ "▁fragments": 26866,
+ "icc": 26867,
+ "▁booth": 26868,
+ "▁cruise": 26869,
+ "▁Testament": 26870,
+ "cola": 26871,
+ "▁Leop": 26872,
+ "▁noon": 26873,
+ "▁terrified": 26874,
+ "vb": 26875,
+ "intel": 26876,
+ "alie": 26877,
+ "▁verification": 26878,
+ "yster": 26879,
+ "ADER": 26880,
+ "chied": 26881,
+ "▁datasets": 26882,
+ "▁зі": 26883,
+ "▁miem": 26884,
+ "ulates": 26885,
+ "▁uuid": 26886,
+ "▁Pictures": 26887,
+ "▁Brend": 26888,
+ "Billboard": 26889,
+ "▁stern": 26890,
+ "▁denom": 26891,
+ "▁accidents": 26892,
+ "сня": 26893,
+ "▁packing": 26894,
+ "ција": 26895,
+ "iblical": 26896,
+ "▁Так": 26897,
+ "▁whisk": 26898,
+ "▁luego": 26899,
+ "▁rectangle": 26900,
+ "▁hooks": 26901,
+ "▁neglect": 26902,
+ "▁sober": 26903,
+ "proposition": 26904,
+ "Multiple": 26905,
+ ":\",": 26906,
+ "▁bapt": 26907,
+ "Parts": 26908,
+ "▁Selection": 26909,
+ "▁Alpha": 26910,
+ "weights": 26911,
+ "hall": 26912,
+ "соб": 26913,
+ "▁lur": 26914,
+ "▁época": 26915,
+ "▁rested": 26916,
+ "ambigu": 26917,
+ "▁tastes": 26918,
+ "amazonaws": 26919,
+ "▁confess": 26920,
+ "▁diciembre": 26921,
+ "implement": 26922,
+ "▁absorption": 26923,
+ "Hal": 26924,
+ "LEAN": 26925,
+ "▁Zach": 26926,
+ "▁freeze": 26927,
+ "LBL": 26928,
+ "STM": 26929,
+ "▁calc": 26930,
+ "={()": 26931,
+ "=*/": 26932,
+ "▁bt": 26933,
+ "Reb": 26934,
+ "▁Wien": 26935,
+ "anska": 26936,
+ "▁surn": 26937,
+ "iative": 26938,
+ "▁invån": 26939,
+ "CY": 26940,
+ "▁là": 26941,
+ "amba": 26942,
+ "leen": 26943,
+ "wahl": 26944,
+ "▁functioning": 26945,
+ "ția": 26946,
+ "getContext": 26947,
+ "gart": 26948,
+ "▁обе": 26949,
+ "Pen": 26950,
+ "vik": 26951,
+ "Slider": 26952,
+ "▁Accept": 26953,
+ "Gap": 26954,
+ "▁Jorge": 26955,
+ "SIG": 26956,
+ "▁вос": 26957,
+ "▁голо": 26958,
+ "▁periodo": 26959,
+ "шта": 26960,
+ "▁patches": 26961,
+ "кої": 26962,
+ "äre": 26963,
+ "engono": 26964,
+ "lista": 26965,
+ "horn": 26966,
+ "▁Complex": 26967,
+ "Sent": 26968,
+ "trfs": 26969,
+ "▁convex": 26970,
+ "Generation": 26971,
+ "▁місце": 26972,
+ "compress": 26973,
+ "▁Sax": 26974,
+ "▁uid": 26975,
+ "▁Lebens": 26976,
+ "Completion": 26977,
+ "\\|_{": 26978,
+ "insky": 26979,
+ "▁schon": 26980,
+ "▁masters": 26981,
+ "independ": 26982,
+ "neys": 26983,
+ "▁lied": 26984,
+ "▁aspir": 26985,
+ "чні": 26986,
+ "▁breakdown": 26987,
+ "▁Harm": 26988,
+ "▁designing": 26989,
+ "hf": 26990,
+ "▁Angela": 26991,
+ "▁confer": 26992,
+ "▁partido": 26993,
+ "▁interference": 26994,
+ "mao": 26995,
+ "▁absorbed": 26996,
+ "▁Vall": 26997,
+ "ErrorCode": 26998,
+ "▁Publishing": 26999,
+ "vano": 27000,
+ "BITS": 27001,
+ "▁deer": 27002,
+ "▁Campaign": 27003,
+ "▁graz": 27004,
+ "CHANGE": 27005,
+ "▁feder": 27006,
+ "iffe": 27007,
+ "handed": 27008,
+ "cq": 27009,
+ "umbing": 27010,
+ "▁unre": 27011,
+ "▁siendo": 27012,
+ "▁simpler": 27013,
+ "why": 27014,
+ "arettes": 27015,
+ "anst": 27016,
+ "▁hass": 27017,
+ "▁Enterprise": 27018,
+ "▁mois": 27019,
+ "▁Fo": 27020,
+ "▁участ": 27021,
+ "ffen": 27022,
+ "▁MODULE": 27023,
+ "▁activated": 27024,
+ "▁internacional": 27025,
+ "▁Mittel": 27026,
+ "degree": 27027,
+ "▁откры": 27028,
+ "▁&(": 27029,
+ "getProperty": 27030,
+ "isz": 27031,
+ "cedure": 27032,
+ "▁enters": 27033,
+ "▁Sally": 27034,
+ "▁Train": 27035,
+ "▁logged": 27036,
+ "▁Rav": 27037,
+ "▁Avoid": 27038,
+ "▁Kaiser": 27039,
+ "▁expend": 27040,
+ "aphor": 27041,
+ "▁brass": 27042,
+ "▁melod": 27043,
+ "▁attitudes": 27044,
+ "*\"": 27045,
+ "Wall": 27046,
+ "▁owe": 27047,
+ "▁bamb": 27048,
+ "shader": 27049,
+ "cester": 27050,
+ "▁PP": 27051,
+ "▁migrations": 27052,
+ "entric": 27053,
+ "▁Setup": 27054,
+ "▁Artist": 27055,
+ "hre": 27056,
+ "▁polite": 27057,
+ "ahan": 27058,
+ "▁luglio": 27059,
+ "▁predecess": 27060,
+ "▁SIG": 27061,
+ "тів": 27062,
+ "▁RF": 27063,
+ "▁Dry": 27064,
+ "▁maker": 27065,
+ "шим": 27066,
+ "▁Sounds": 27067,
+ "▁implementing": 27068,
+ "▁ah": 27069,
+ "▁gev": 27070,
+ "▁duplicate": 27071,
+ "▁Logan": 27072,
+ "▁Grade": 27073,
+ "DUCT": 27074,
+ "íses": 27075,
+ "ért": 27076,
+ "▁nonsense": 27077,
+ "backup": 27078,
+ "Attachment": 27079,
+ "▁ecc": 27080,
+ "▁Squadron": 27081,
+ "learn": 27082,
+ "deprecated": 27083,
+ "▁Aub": 27084,
+ "▁Gol": 27085,
+ "▁overl": 27086,
+ "SERVICE": 27087,
+ "▁beautifully": 27088,
+ "REL": 27089,
+ "▁Gian": 27090,
+ "▁Papa": 27091,
+ "respond": 27092,
+ "▁Caribbean": 27093,
+ "rn": 27094,
+ "▁худож": 27095,
+ "Cfg": 27096,
+ "rai": 27097,
+ "▁sniff": 27098,
+ "tto": 27099,
+ "ологи": 27100,
+ "▁rb": 27101,
+ "▁incidents": 27102,
+ "▁duck": 27103,
+ "▁PROVIDED": 27104,
+ "Sources": 27105,
+ "▁Chelsea": 27106,
+ "▁tek": 27107,
+ "▁налази": 27108,
+ "▁pilots": 27109,
+ "тки": 27110,
+ "▁traded": 27111,
+ "▁Beijing": 27112,
+ "▁Gregory": 27113,
+ "scalar": 27114,
+ "▁inclined": 27115,
+ "▁Kamp": 27116,
+ "▁Marian": 27117,
+ "▁fierce": 27118,
+ "▁theft": 27119,
+ "ющих": 27120,
+ "▁Into": 27121,
+ "constraint": 27122,
+ "parentNode": 27123,
+ "idental": 27124,
+ "▁gouvernement": 27125,
+ "▁SND": 27126,
+ "▁Ruby": 27127,
+ "▁monaster": 27128,
+ "Records": 27129,
+ "▁Kab": 27130,
+ "▁Universe": 27131,
+ "▁approximate": 27132,
+ "Water": 27133,
+ "▁Physical": 27134,
+ "appers": 27135,
+ "oubtedly": 27136,
+ "ложен": 27137,
+ "▁towel": 27138,
+ "▁siblings": 27139,
+ "eph": 27140,
+ "icios": 27141,
+ "рами": 27142,
+ "▁outrage": 27143,
+ "▁també": 27144,
+ "SRC": 27145,
+ "телем": 27146,
+ "Vi": 27147,
+ ".');": 27148,
+ "LM": 27149,
+ "▁mitt": 27150,
+ "▁weed": 27151,
+ "▁crops": 27152,
+ "iman": 27153,
+ "Claim": 27154,
+ "insula": 27155,
+ "▁(“": 27156,
+ "▁Changes": 27157,
+ "▁invånare": 27158,
+ "again": 27159,
+ "▁cnt": 27160,
+ "▁Gaz": 27161,
+ "▁austral": 27162,
+ "overlay": 27163,
+ "▁Mechan": 27164,
+ "▁slammed": 27165,
+ "▁trailing": 27166,
+ "▁Biography": 27167,
+ "▁appealing": 27168,
+ "IVER": 27169,
+ "▁Ave": 27170,
+ "▁Plot": 27171,
+ "voj": 27172,
+ "▁sung": 27173,
+ "▁unos": 27174,
+ "Effects": 27175,
+ "vv": 27176,
+ "cook": 27177,
+ "Buttons": 27178,
+ "▁transm": 27179,
+ "ierto": 27180,
+ "CONTEXT": 27181,
+ "▁dignity": 27182,
+ "aired": 27183,
+ "javax": 27184,
+ "▁Alberto": 27185,
+ "▁Recently": 27186,
+ "▁facial": 27187,
+ "mathop": 27188,
+ "ało": 27189,
+ "вид": 27190,
+ "cott": 27191,
+ "Variables": 27192,
+ "▁Ran": 27193,
+ "▁bunk": 27194,
+ "amiliar": 27195,
+ "CAST": 27196,
+ "▁frü": 27197,
+ "VED": 27198,
+ "▁NOTICE": 27199,
+ "▁turno": 27200,
+ "validator": 27201,
+ "▁Portuguese": 27202,
+ "▁questioning": 27203,
+ "}})": 27204,
+ "▁lear": 27205,
+ "Xamarin": 27206,
+ "▁disadv": 27207,
+ "encoded": 27208,
+ "▁Kot": 27209,
+ "rated": 27210,
+ "▁Theory": 27211,
+ "cius": 27212,
+ "▁Darwin": 27213,
+ "ђе": 27214,
+ "▁décl": 27215,
+ "▁область": 27216,
+ "рович": 27217,
+ "▁mobility": 27218,
+ "VF": 27219,
+ "▁хи": 27220,
+ "until": 27221,
+ "▁barriers": 27222,
+ "gif": 27223,
+ "▁Roh": 27224,
+ "▁aging": 27225,
+ "▁Widget": 27226,
+ "olk": 27227,
+ "▁farms": 27228,
+ "Checker": 27229,
+ "Introduction": 27230,
+ "смо": 27231,
+ "▁Russians": 27232,
+ "naments": 27233,
+ "▁Insert": 27234,
+ "▁Whenever": 27235,
+ "erset": 27236,
+ "itori": 27237,
+ "▁Dort": 27238,
+ "▁costume": 27239,
+ "▁mathematical": 27240,
+ "▁Bast": 27241,
+ "▁nominated": 27242,
+ "▁restoration": 27243,
+ "posal": 27244,
+ "▁unfortunate": 27245,
+ "Ps": 27246,
+ "LIN": 27247,
+ "▁intact": 27248,
+ "▁provoc": 27249,
+ "▁située": 27250,
+ "▁ноября": 27251,
+ "ermo": 27252,
+ "▁fisher": 27253,
+ "гля": 27254,
+ "▁conting": 27255,
+ "▁Doug": 27256,
+ "\"?": 27257,
+ "▁Eva": 27258,
+ "▁tops": 27259,
+ "▁Remote": 27260,
+ "▁artwork": 27261,
+ "▁artillery": 27262,
+ "quick": 27263,
+ "▁Arabia": 27264,
+ "▁SDValue": 27265,
+ "▁Dakota": 27266,
+ "iated": 27267,
+ "▁Optim": 27268,
+ "buttons": 27269,
+ "▁cottage": 27270,
+ "▁wherein": 27271,
+ "▁tutorial": 27272,
+ "▁Scre": 27273,
+ "▁sweep": 27274,
+ "▁Coffee": 27275,
+ "})}": 27276,
+ "▁музы": 27277,
+ "hostname": 27278,
+ "▁Temp": 27279,
+ "▁Fut": 27280,
+ "respect": 27281,
+ "ocz": 27282,
+ "▁predomin": 27283,
+ "Indicator": 27284,
+ "encial": 27285,
+ "UMENT": 27286,
+ "▁SHALL": 27287,
+ "▁commanded": 27288,
+ "▁withdrawal": 27289,
+ "iour": 27290,
+ "REGION": 27291,
+ "sprintf": 27292,
+ "▁вме": 27293,
+ "▁Payment": 27294,
+ "▁Anim": 27295,
+ "publish": 27296,
+ "▁seeks": 27297,
+ "ouw": 27298,
+ "▁GM": 27299,
+ "rugu": 27300,
+ "ustain": 27301,
+ "▁))": 27302,
+ "▁consulting": 27303,
+ "▁Dialog": 27304,
+ "▁Lars": 27305,
+ "▁critique": 27306,
+ "▁circulation": 27307,
+ "▁landsc": 27308,
+ "managed": 27309,
+ "▁Craft": 27310,
+ "▁herman": 27311,
+ "afi": 27312,
+ "amy": 27313,
+ "▁discour": 27314,
+ "<>(": 27315,
+ "▁Steph": 27316,
+ "▁tolerance": 27317,
+ "typename": 27318,
+ "ventions": 27319,
+ "ział": 27320,
+ "стов": 27321,
+ "▁sticking": 27322,
+ "ASC": 27323,
+ "ISO": 27324,
+ "▁Spencer": 27325,
+ "▁Didn": 27326,
+ "gomery": 27327,
+ "imiter": 27328,
+ "dru": 27329,
+ "Clause": 27330,
+ "▁slides": 27331,
+ "###": 27332,
+ "▁Sugar": 27333,
+ "HY": 27334,
+ "▁эти": 27335,
+ "▁Edwards": 27336,
+ "▁cents": 27337,
+ "oya": 27338,
+ "serts": 27339,
+ "▁Hass": 27340,
+ "▁ingen": 27341,
+ "стри": 27342,
+ "▁saddle": 27343,
+ "solid": 27344,
+ "▁champions": 27345,
+ "-)": 27346,
+ "▁Slov": 27347,
+ "▁shiny": 27348,
+ "▁*)&": 27349,
+ "▁Define": 27350,
+ "če": 27351,
+ "▁scrut": 27352,
+ "onden": 27353,
+ "'\",": 27354,
+ "uffs": 27355,
+ "▁olymp": 27356,
+ "idential": 27357,
+ "wand": 27358,
+ "▁annually": 27359,
+ "▁Arkansas": 27360,
+ "▁saint": 27361,
+ "▁gleich": 27362,
+ "▁perfection": 27363,
+ ")>": 27364,
+ "▁shorts": 27365,
+ "▁justified": 27366,
+ "peated": 27367,
+ "packages": 27368,
+ "driven": 27369,
+ "▁Liberty": 27370,
+ "▁stripped": 27371,
+ "шение": 27372,
+ "▁fünf": 27373,
+ "▁ecosystem": 27374,
+ "ixa": 27375,
+ "▁Fresh": 27376,
+ "vart": 27377,
+ "▁treats": 27378,
+ "▁stance": 27379,
+ "чёт": 27380,
+ "▁pity": 27381,
+ "adém": 27382,
+ "▁окон": 27383,
+ "▁Chand": 27384,
+ "rab": 27385,
+ "вший": 27386,
+ "inski": 27387,
+ "▁continually": 27388,
+ "▁Daddy": 27389,
+ "▁nightmare": 27390,
+ "icional": 27391,
+ "▁efect": 27392,
+ "ueblo": 27393,
+ "▁lanç": 27394,
+ "▁Collections": 27395,
+ "due": 27396,
+ "ampton": 27397,
+ "▁memcpy": 27398,
+ "▁**(": 27399,
+ "issent": 27400,
+ "▁Insp": 27401,
+ "▁Glasgow": 27402,
+ "▁furono": 27403,
+ "▁kindness": 27404,
+ "Bi": 27405,
+ "▁competed": 27406,
+ "▁oak": 27407,
+ "Large": 27408,
+ "▁disgu": 27409,
+ "▁kings": 27410,
+ "тами": 27411,
+ "▁stuffed": 27412,
+ "▁hilar": 27413,
+ "published": 27414,
+ "▁stressed": 27415,
+ "▁Peak": 27416,
+ "▁loader": 27417,
+ "Keyboard": 27418,
+ "▁reconstruction": 27419,
+ "▁vod": 27420,
+ "▁dun": 27421,
+ "▁understands": 27422,
+ "tenant": 27423,
+ "▁chaque": 27424,
+ "▁prejud": 27425,
+ "utat": 27426,
+ "▁uso": 27427,
+ "▁Heavy": 27428,
+ "▁cuatro": 27429,
+ "▁sidewalk": 27430,
+ "▁Bug": 27431,
+ "▁månaden": 27432,
+ "geo": 27433,
+ "▁united": 27434,
+ "▁Files": 27435,
+ "▁Аль": 27436,
+ "▁rugby": 27437,
+ "▁financing": 27438,
+ "▁comply": 27439,
+ "": 27440,
+ "▁rushing": 27441,
+ "▁fen": 27442,
+ "mong": 27443,
+ "▁spé": 27444,
+ "▁presenting": 27445,
+ "INCLUDING": 27446,
+ "ěl": 27447,
+ "zeichnung": 27448,
+ "Backup": 27449,
+ "▁petit": 27450,
+ "▁allerg": 27451,
+ "нут": 27452,
+ "▁worrying": 27453,
+ "▁mamm": 27454,
+ "▁operand": 27455,
+ ":%.*]]": 27456,
+ "▁realise": 27457,
+ "Commands": 27458,
+ "▁Bew": 27459,
+ "▁assumes": 27460,
+ "▁Covid": 27461,
+ "▁quand": 27462,
+ "tyard": 27463,
+ "▁Mono": 27464,
+ "linked": 27465,
+ "MARK": 27466,
+ "Esp": 27467,
+ "▁blessing": 27468,
+ "▁eyebrows": 27469,
+ "▁NV": 27470,
+ "▁стру": 27471,
+ "▁modeling": 27472,
+ "▁greeted": 27473,
+ "Workspace": 27474,
+ "▁pedest": 27475,
+ "▁неза": 27476,
+ "lemagne": 27477,
+ "Statistics": 27478,
+ "▁aument": 27479,
+ "▁speeds": 27480,
+ "▁syndrome": 27481,
+ "CONNECT": 27482,
+ "zahl": 27483,
+ "verso": 27484,
+ "ército": 27485,
+ "▁astronom": 27486,
+ "▁aprile": 27487,
+ "žen": 27488,
+ "веро": 27489,
+ "draft": 27490,
+ "▁gioc": 27491,
+ "▁comport": 27492,
+ "▁variance": 27493,
+ "▁realizing": 27494,
+ "EDIT": 27495,
+ "олові": 27496,
+ "▁estar": 27497,
+ "▁sost": 27498,
+ "NORMAL": 27499,
+ "▁ó": 27500,
+ "▁Andr": 27501,
+ "ATTRIB": 27502,
+ "▁rede": 27503,
+ "▁toes": 27504,
+ "▁advances": 27505,
+ "▁Against": 27506,
+ "TOM": 27507,
+ "rss": 27508,
+ "MMMM": 27509,
+ "▁newest": 27510,
+ "▁VER": 27511,
+ "▁phrases": 27512,
+ "anter": 27513,
+ "Launch": 27514,
+ "▁chr": 27515,
+ "▁manufactured": 27516,
+ "$),": 27517,
+ "rollment": 27518,
+ "eston": 27519,
+ "▁peint": 27520,
+ "”)": 27521,
+ "endet": 27522,
+ "▁Hair": 27523,
+ "ivalent": 27524,
+ "▁upright": 27525,
+ "gren": 27526,
+ "anked": 27527,
+ "wright": 27528,
+ "▁mast": 27529,
+ "▁onChange": 27530,
+ "▁debris": 27531,
+ "▁grap": 27532,
+ "etry": 27533,
+ "▁(__": 27534,
+ "▁Commerce": 27535,
+ "BOX": 27536,
+ "Tax": 27537,
+ "▁отри": 27538,
+ "▁prevention": 27539,
+ "▁Feel": 27540,
+ "▁exotic": 27541,
+ "▁Bark": 27542,
+ "▁Steam": 27543,
+ "fon": 27544,
+ "olin": 27545,
+ "▁eliminated": 27546,
+ "▁bc": 27547,
+ "▁Cycl": 27548,
+ "▁$(\"#": 27549,
+ "▁Parl": 27550,
+ "manuel": 27551,
+ "ospher": 27552,
+ "WF": 27553,
+ "Analy": 27554,
+ "▁navig": 27555,
+ "▁renown": 27556,
+ "Rx": 27557,
+ "▁Walt": 27558,
+ "uffed": 27559,
+ "▁foster": 27560,
+ "$:": 27561,
+ "shore": 27562,
+ "Connector": 27563,
+ "фика": 27564,
+ "▁realization": 27565,
+ "Li": 27566,
+ "ctxt": 27567,
+ "ahoo": 27568,
+ "▁miracle": 27569,
+ "▁ET": 27570,
+ "▁GPS": 27571,
+ "▁Observable": 27572,
+ "▁hf": 27573,
+ "▁magnificent": 27574,
+ "него": 27575,
+ "BIN": 27576,
+ "▁Dorf": 27577,
+ "ieck": 27578,
+ "vee": 27579,
+ "▁Craw": 27580,
+ "/#": 27581,
+ "▁pci": 27582,
+ "ippet": 27583,
+ "▁Hillary": 27584,
+ "▁gir": 27585,
+ "▁rand": 27586,
+ "▁laying": 27587,
+ "▁Different": 27588,
+ "boys": 27589,
+ "virt": 27590,
+ "▁encryption": 27591,
+ "ász": 27592,
+ "пор": 27593,
+ "▁smelled": 27594,
+ "▁suscept": 27595,
+ "cluded": 27596,
+ "▁Carn": 27597,
+ "igten": 27598,
+ "▁Chuck": 27599,
+ "▁Provinc": 27600,
+ "▁perí": 27601,
+ "▁Marshal": 27602,
+ "мож": 27603,
+ "gfx": 27604,
+ "oshi": 27605,
+ "▁WHE": 27606,
+ "▁relaxation": 27607,
+ ",.": 27608,
+ "were": 27609,
+ "▁varieties": 27610,
+ "▁Won": 27611,
+ "▁gaps": 27612,
+ "▁stole": 27613,
+ "igua": 27614,
+ "ющие": 27615,
+ "▁Hampshire": 27616,
+ "phrase": 27617,
+ "▁película": 27618,
+ "Processing": 27619,
+ "▁initialization": 27620,
+ "oustic": 27621,
+ "▁Josef": 27622,
+ "icating": 27623,
+ "▁goodness": 27624,
+ "TES": 27625,
+ "▁cope": 27626,
+ "▁ignorance": 27627,
+ "▁Brist": 27628,
+ "▁paras": 27629,
+ "▁accidentally": 27630,
+ "▁tand": 27631,
+ "ittest": 27632,
+ "▁ули": 27633,
+ "▁shipped": 27634,
+ "▁ост": 27635,
+ "elseif": 27636,
+ "▁usize": 27637,
+ "horizontal": 27638,
+ "▁Carr": 27639,
+ "▁precip": 27640,
+ "roz": 27641,
+ "pathetic": 27642,
+ "rived": 27643,
+ "rok": 27644,
+ "▁digging": 27645,
+ "мом": 27646,
+ "▁Mull": 27647,
+ "▁XIII": 27648,
+ "▁peas": 27649,
+ "▁foul": 27650,
+ "▁travels": 27651,
+ "▁Ng": 27652,
+ "▁составе": 27653,
+ "Mont": 27654,
+ "arde": 27655,
+ "▁Stefan": 27656,
+ "^^^^": 27657,
+ "▁Kiss": 27658,
+ "▁Ek": 27659,
+ "▁oktober": 27660,
+ "▁memorable": 27661,
+ "')).": 27662,
+ "▁Vision": 27663,
+ "▁Nina": 27664,
+ "▁Solar": 27665,
+ "▁highlighted": 27666,
+ "▁memo": 27667,
+ "meisterschaft": 27668,
+ "sidebar": 27669,
+ "SEE": 27670,
+ "▁Nevada": 27671,
+ "Da": 27672,
+ "▁drawer": 27673,
+ "astically": 27674,
+ "elde": 27675,
+ "scribed": 27676,
+ "▁priests": 27677,
+ "▁hommes": 27678,
+ "▁instructor": 27679,
+ "клад": 27680,
+ "▁spett": 27681,
+ "\\-": 27682,
+ "▁мира": 27683,
+ "▁Looks": 27684,
+ "▁sleeve": 27685,
+ "▁strongest": 27686,
+ "▁tête": 27687,
+ "▁Nicole": 27688,
+ "imper": 27689,
+ "нача": 27690,
+ "ipper": 27691,
+ "▁inwon": 27692,
+ "ilers": 27693,
+ "▁Deputy": 27694,
+ "oge": 27695,
+ "▁depressed": 27696,
+ "▁arte": 27697,
+ "▁combining": 27698,
+ "LAST": 27699,
+ "inted": 27700,
+ "▁Average": 27701,
+ "▁pollution": 27702,
+ "▁Phillips": 27703,
+ "▁WM": 27704,
+ "}}}\\": 27705,
+ "Added": 27706,
+ "▁peripher": 27707,
+ "Creation": 27708,
+ "▁italien": 27709,
+ "▁Choice": 27710,
+ "▁EXPRESS": 27711,
+ "▁Struct": 27712,
+ "ysz": 27713,
+ "Resize": 27714,
+ "ARGS": 27715,
+ "▁repo": 27716,
+ "▁чтобы": 27717,
+ "▁pref": 27718,
+ "▁earthqu": 27719,
+ "▁Мекси": 27720,
+ "▁Finale": 27721,
+ "▁hecho": 27722,
+ "requests": 27723,
+ "Cut": 27724,
+ "▁deserved": 27725,
+ "гово": 27726,
+ "▁Recent": 27727,
+ "▁дивизи": 27728,
+ "▁supportive": 27729,
+ "прави": 27730,
+ "▁irrelevant": 27731,
+ "'\r": 27732,
+ "▁ctrl": 27733,
+ "▁Deal": 27734,
+ "izada": 27735,
+ "uo": 27736,
+ "▁nort": 27737,
+ "geometry": 27738,
+ "▁Individual": 27739,
+ "ereg": 27740,
+ "▁приня": 27741,
+ "cref": 27742,
+ "══": 27743,
+ "▁comerc": 27744,
+ "=_": 27745,
+ "bund": 27746,
+ "тах": 27747,
+ "ilen": 27748,
+ "чита": 27749,
+ "▁corporation": 27750,
+ "esz": 27751,
+ "▁==>": 27752,
+ "ablish": 27753,
+ "Apr": 27754,
+ "▁ripped": 27755,
+ "Vars": 27756,
+ "stret": 27757,
+ "▁Francesco": 27758,
+ "NaN": 27759,
+ "▁anytime": 27760,
+ "▁automated": 27761,
+ "ostream": 27762,
+ "▁drawings": 27763,
+ "▁enhancement": 27764,
+ "okrat": 27765,
+ "▁Issue": 27766,
+ "вра": 27767,
+ "Currency": 27768,
+ "▁wyn": 27769,
+ "izarre": 27770,
+ "ético": 27771,
+ "multiple": 27772,
+ "▁Rate": 27773,
+ "▁Ich": 27774,
+ "▁Auss": 27775,
+ "▁Former": 27776,
+ "Curve": 27777,
+ "▁marvel": 27778,
+ "attro": 27779,
+ "▁сп": 27780,
+ "BOOL": 27781,
+ "сия": 27782,
+ "gold": 27783,
+ "▁Nintendo": 27784,
+ "▁Salvador": 27785,
+ "▁Solution": 27786,
+ "ADC": 27787,
+ "бора": 27788,
+ "▁Bennett": 27789,
+ "▁FR": 27790,
+ "▁pueden": 27791,
+ "patient": 27792,
+ "▁PG": 27793,
+ "▁Jin": 27794,
+ "▁crashed": 27795,
+ "▁denen": 27796,
+ "▁Sample": 27797,
+ "▁Quebec": 27798,
+ "itories": 27799,
+ "▁blinked": 27800,
+ "▁lion": 27801,
+ "▁voce": 27802,
+ "▁Impact": 27803,
+ "▁Mau": 27804,
+ "▁Nie": 27805,
+ "▁lob": 27806,
+ "▁две": 27807,
+ "orneys": 27808,
+ "▁coastal": 27809,
+ "▁sensors": 27810,
+ "▁XII": 27811,
+ "▁illusion": 27812,
+ "oji": 27813,
+ "▁INC": 27814,
+ "▁Duncan": 27815,
+ "yk": 27816,
+ "▁affecting": 27817,
+ "pul": 27818,
+ "▁Napoleon": 27819,
+ "▁акаде": 27820,
+ "▁compt": 27821,
+ "▁profitable": 27822,
+ "loe": 27823,
+ "▁deuxième": 27824,
+ "▁WC": 27825,
+ "▁viable": 27826,
+ "▁Drug": 27827,
+ "TextBox": 27828,
+ "▁luminos": 27829,
+ "auté": 27830,
+ "yc": 27831,
+ "ště": 27832,
+ "▁affiliates": 27833,
+ "ilda": 27834,
+ "conduct": 27835,
+ "▁ebenfalls": 27836,
+ "▁AMD": 27837,
+ "▁Monitor": 27838,
+ "▁Companies": 27839,
+ "▁corrected": 27840,
+ "äck": 27841,
+ "SYSTEM": 27842,
+ "otherapy": 27843,
+ "▁перед": 27844,
+ "▁blues": 27845,
+ "atisf": 27846,
+ "although": 27847,
+ "rost": 27848,
+ "SCAN": 27849,
+ "▁RAM": 27850,
+ "ціональ": 27851,
+ "▁vendors": 27852,
+ "▁customs": 27853,
+ "▁activate": 27854,
+ "▁blogs": 27855,
+ "▁brace": 27856,
+ "▁strat": 27857,
+ "anje": 27858,
+ "щё": 27859,
+ "▁tide": 27860,
+ "▁Brigade": 27861,
+ "getOperand": 27862,
+ "▁aliment": 27863,
+ "▁achievements": 27864,
+ "▁suspicion": 27865,
+ "▁touchdown": 27866,
+ "broad": 27867,
+ "iore": 27868,
+ "Comparison": 27869,
+ "▁mum": 27870,
+ "English": 27871,
+ "▁Picture": 27872,
+ "▁Mouse": 27873,
+ "amd": 27874,
+ "▁[`": 27875,
+ "▁denomin": 27876,
+ "▁Aleks": 27877,
+ "▁prevents": 27878,
+ "ób": 27879,
+ "fed": 27880,
+ "▁Pray": 27881,
+ "▁shine": 27882,
+ "▁clutch": 27883,
+ "mux": 27884,
+ "Appro": 27885,
+ "▁notably": 27886,
+ "chio": 27887,
+ "nage": 27888,
+ "HAS": 27889,
+ "▁')": 27890,
+ "▁Miche": 27891,
+ "tg": 27892,
+ "::~": 27893,
+ "▁amely": 27894,
+ "▁rodz": 27895,
+ "zs": 27896,
+ "trait": 27897,
+ "▁klass": 27898,
+ "fö": 27899,
+ "▁destac": 27900,
+ "▁Clara": 27901,
+ "frequency": 27902,
+ "▁Git": 27903,
+ "▁поль": 27904,
+ "▁frequencies": 27905,
+ "▁febrero": 27906,
+ "▁stumbled": 27907,
+ "кою": 27908,
+ "▁Names": 27909,
+ "▁Flight": 27910,
+ "▁prey": 27911,
+ "▁medio": 27912,
+ "▁VAR": 27913,
+ "▁Float": 27914,
+ "▁Ernest": 27915,
+ "▁Marcatori": 27916,
+ "oport": 27917,
+ "▁cancellation": 27918,
+ "▁Bryan": 27919,
+ "————": 27920,
+ "Luc": 27921,
+ "▁libre": 27922,
+ "▁título": 27923,
+ "*>": 27924,
+ "▁Sandy": 27925,
+ "▁Marina": 27926,
+ "Been": 27927,
+ "▁wal": 27928,
+ "▁Kultur": 27929,
+ "▁explode": 27930,
+ "▁limiting": 27931,
+ "▁presumably": 27932,
+ "▁pb": 27933,
+ "▁Merc": 27934,
+ "▁реки": 27935,
+ "learning": 27936,
+ "Catalog": 27937,
+ "▁Census": 27938,
+ "lte": 27939,
+ "▁NET": 27940,
+ "raising": 27941,
+ "ське": 27942,
+ "staff": 27943,
+ "▁Quinn": 27944,
+ "▁memorial": 27945,
+ "пня": 27946,
+ "▁cuenta": 27947,
+ "▁XI": 27948,
+ "lbl": 27949,
+ "▁varies": 27950,
+ "▁fluctuations": 27951,
+ "▁долж": 27952,
+ "▁особи": 27953,
+ "▁warehouse": 27954,
+ "However": 27955,
+ "▁corrections": 27956,
+ "dhd": 27957,
+ "▁fals": 27958,
+ "▁controversy": 27959,
+ "▁curse": 27960,
+ "▁télé": 27961,
+ "řed": 27962,
+ "▁AU": 27963,
+ "▁тор": 27964,
+ "▁crít": 27965,
+ "idan": 27966,
+ "iliary": 27967,
+ "▁Panel": 27968,
+ "cule": 27969,
+ "▁Poor": 27970,
+ "▁BA": 27971,
+ "▁ignorant": 27972,
+ "èmes": 27973,
+ "▁aesthetic": 27974,
+ "Linked": 27975,
+ "getInt": 27976,
+ "Unicode": 27977,
+ "[@": 27978,
+ "▁Zent": 27979,
+ "Manifest": 27980,
+ "▁vars": 27981,
+ "PB": 27982,
+ "▁ву": 27983,
+ "▁Describe": 27984,
+ "▁Anything": 27985,
+ "oirs": 27986,
+ "▁socks": 27987,
+ "▁imped": 27988,
+ "▁neue": 27989,
+ "▁dispers": 27990,
+ "Collect": 27991,
+ "filer": 27992,
+ "▁Frau": 27993,
+ "▁Hockey": 27994,
+ "▁teens": 27995,
+ "▁Roberto": 27996,
+ "lauf": 27997,
+ "вать": 27998,
+ "▁ско": 27999,
+ "isArray": 28000,
+ "▁teenager": 28001,
+ "Built": 28002,
+ "▁loudly": 28003,
+ "Capacity": 28004,
+ "▁adventures": 28005,
+ "▁Molly": 28006,
+ "recogn": 28007,
+ "bars": 28008,
+ "▁Lor": 28009,
+ "▁può": 28010,
+ "▁mong": 28011,
+ "inement": 28012,
+ "Assignment": 28013,
+ "▁diz": 28014,
+ "lessness": 28015,
+ "▁Halloween": 28016,
+ "▁bitmap": 28017,
+ "Rom": 28018,
+ "нар": 28019,
+ "▁rebel": 28020,
+ "▁radial": 28021,
+ "measure": 28022,
+ "nit": 28023,
+ "▁Assume": 28024,
+ "▁assignments": 28025,
+ "▁Isn": 28026,
+ "▁altre": 28027,
+ "ßer": 28028,
+ "наль": 28029,
+ "▁flies": 28030,
+ "▁droit": 28031,
+ "▁thickness": 28032,
+ "▁enjo": 28033,
+ "▁dwell": 28034,
+ "▁homosexual": 28035,
+ "▁eval": 28036,
+ "$_{": 28037,
+ "asia": 28038,
+ "▁philos": 28039,
+ "getCurrent": 28040,
+ "▁veterans": 28041,
+ "▁Berkeley": 28042,
+ "▁wildlife": 28043,
+ "Cop": 28044,
+ "vern": 28045,
+ "▁Ú": 28046,
+ "tos": 28047,
+ "▁Led": 28048,
+ "▁keywords": 28049,
+ "▁medications": 28050,
+ "neum": 28051,
+ "▁jamais": 28052,
+ "▁Buc": 28053,
+ "▁PD": 28054,
+ "▁Statement": 28055,
+ "▁PI": 28056,
+ "▁Jackie": 28057,
+ "▁ordin": 28058,
+ "▁kör": 28059,
+ "enze": 28060,
+ "▁utilized": 28061,
+ "áct": 28062,
+ "azed": 28063,
+ "▁severely": 28064,
+ "▁även": 28065,
+ "▁libro": 28066,
+ "▁Eu": 28067,
+ "äst": 28068,
+ "PART": 28069,
+ "▁Butler": 28070,
+ "▁puzzle": 28071,
+ "Fall": 28072,
+ "Country": 28073,
+ "pfn": 28074,
+ "▁україн": 28075,
+ "▁Orchestra": 28076,
+ "▁alto": 28077,
+ "▁ancora": 28078,
+ "▁decomposition": 28079,
+ "▁م": 28080,
+ "▁appetite": 28081,
+ "adu": 28082,
+ "▁THAT": 28083,
+ "▁comenz": 28084,
+ "mina": 28085,
+ "▁initiated": 28086,
+ "▁Tat": 28087,
+ "▁sometime": 28088,
+ "rek": 28089,
+ "bread": 28090,
+ "▁Statistics": 28091,
+ "▁Cob": 28092,
+ "Follow": 28093,
+ "▁geometric": 28094,
+ "шла": 28095,
+ "▁proceedings": 28096,
+ "Dlg": 28097,
+ "seven": 28098,
+ "▁[-": 28099,
+ "▁Buffalo": 28100,
+ "▁blacks": 28101,
+ "▁sov": 28102,
+ "▁custody": 28103,
+ "▁ras": 28104,
+ "▁tattoo": 28105,
+ "öffentlicht": 28106,
+ "Blo": 28107,
+ "Austral": 28108,
+ "▁recuper": 28109,
+ "лев": 28110,
+ "▁bem": 28111,
+ "▁thou": 28112,
+ "oriented": 28113,
+ "vir": 28114,
+ "▁colony": 28115,
+ "▁Stanford": 28116,
+ "Absolute": 28117,
+ "adrat": 28118,
+ "▁Situ": 28119,
+ "▁souvent": 28120,
+ "EXEC": 28121,
+ "▁mű": 28122,
+ "▁apartments": 28123,
+ "▁случа": 28124,
+ "▁ano": 28125,
+ "WINDO": 28126,
+ "acci": 28127,
+ "▁Lau": 28128,
+ "court": 28129,
+ "▁manifold": 28130,
+ "▁coalition": 28131,
+ "▁XIV": 28132,
+ "Attrib": 28133,
+ "ascade": 28134,
+ "▁wheat": 28135,
+ "▁strengths": 28136,
+ "FREE": 28137,
+ "EMPTY": 28138,
+ "▁hey": 28139,
+ "ascular": 28140,
+ "▁plasma": 28141,
+ "▁bob": 28142,
+ "Separator": 28143,
+ "=\"${": 28144,
+ "▁Zag": 28145,
+ "▁projet": 28146,
+ "▁smoothly": 28147,
+ "SEQU": 28148,
+ "analy": 28149,
+ "attachment": 28150,
+ "▁ES": 28151,
+ "▁popped": 28152,
+ "ős": 28153,
+ "tom": 28154,
+ "▁són": 28155,
+ "▁rott": 28156,
+ "Utilities": 28157,
+ "hadoop": 28158,
+ "▁sotto": 28159,
+ "autor": 28160,
+ "▁Georges": 28161,
+ "▁který": 28162,
+ "▁gruppo": 28163,
+ "▁когда": 28164,
+ "▁меда": 28165,
+ "▁instrumental": 28166,
+ "▁Writer": 28167,
+ "▁setTimeout": 28168,
+ "ikk": 28169,
+ "▁Dopo": 28170,
+ "]);\r": 28171,
+ "▁practicing": 28172,
+ "▁Ronald": 28173,
+ "▁уби": 28174,
+ "▁agrees": 28175,
+ "▁denoted": 28176,
+ "ismiss": 28177,
+ "▁interviewed": 28178,
+ "templates": 28179,
+ "ři": 28180,
+ "administr": 28181,
+ "▁Butter": 28182,
+ "▁XVII": 28183,
+ "▁positioned": 28184,
+ "▁Fourth": 28185,
+ "▁overwhelmed": 28186,
+ "▁Regular": 28187,
+ "▁reprezent": 28188,
+ "кономи": 28189,
+ "▁expects": 28190,
+ "Indices": 28191,
+ "▁marijuana": 28192,
+ "▁zaj": 28193,
+ "▁Bren": 28194,
+ "▁begg": 28195,
+ "▁nahm": 28196,
+ "▁interrog": 28197,
+ "тие": 28198,
+ "▁Bun": 28199,
+ "▁серед": 28200,
+ "▁shelves": 28201,
+ "▁которых": 28202,
+ "▁Frauen": 28203,
+ "▁Sergeant": 28204,
+ "▁успе": 28205,
+ "matched": 28206,
+ "▁donne": 28207,
+ "▁touches": 28208,
+ "abort": 28209,
+ "▁vale": 28210,
+ "▁institutional": 28211,
+ "▁Mons": 28212,
+ "▁ambitious": 28213,
+ "▁nonetheless": 28214,
+ "jd": 28215,
+ "пей": 28216,
+ "▁backpack": 28217,
+ "dao": 28218,
+ "вия": 28219,
+ "▁surroundings": 28220,
+ "|_{": 28221,
+ "▁gegründ": 28222,
+ "disp": 28223,
+ "▁moisture": 28224,
+ "▁wyd": 28225,
+ "▁traders": 28226,
+ "▁Erst": 28227,
+ "▁Galaxy": 28228,
+ "▁воло": 28229,
+ "▁Peru": 28230,
+ "▁priorities": 28231,
+ "▁pronounced": 28232,
+ "▁CBS": 28233,
+ "▁Palm": 28234,
+ "▁expans": 28235,
+ "▁energet": 28236,
+ "▁Condition": 28237,
+ "▁Sver": 28238,
+ "nested": 28239,
+ "▁февраля": 28240,
+ "hero": 28241,
+ "▁коло": 28242,
+ "▁Films": 28243,
+ "Bon": 28244,
+ "éal": 28245,
+ "ployed": 28246,
+ "trained": 28247,
+ "▁első": 28248,
+ "▁lust": 28249,
+ "atinum": 28250,
+ "oyle": 28251,
+ "▁Jet": 28252,
+ "ждения": 28253,
+ "▁surveys": 28254,
+ "bee": 28255,
+ "workers": 28256,
+ "records": 28257,
+ "calendar": 28258,
+ "bbing": 28259,
+ "regation": 28260,
+ "dashboard": 28261,
+ "King": 28262,
+ "▁vista": 28263,
+ "▁depicted": 28264,
+ "▁occurring": 28265,
+ "▁офи": 28266,
+ "▁sandwich": 28267,
+ "rcu": 28268,
+ "kern": 28269,
+ "▁minut": 28270,
+ "▁смер": 28271,
+ "▁td": 28272,
+ "solete": 28273,
+ "Complex": 28274,
+ "▁tunn": 28275,
+ "▁scarc": 28276,
+ "stead": 28277,
+ "▁Fail": 28278,
+ "▁Rs": 28279,
+ "▁trails": 28280,
+ "kem": 28281,
+ "▁Romans": 28282,
+ "ativity": 28283,
+ "Previous": 28284,
+ "▁depress": 28285,
+ "▁resigned": 28286,
+ "getDefault": 28287,
+ "▁Tibet": 28288,
+ "▁Franco": 28289,
+ "\")));": 28290,
+ "▁injection": 28291,
+ "removed": 28292,
+ "▁praised": 28293,
+ "▁Asc": 28294,
+ "erase": 28295,
+ "▁commissioned": 28296,
+ "MAIL": 28297,
+ "▁Boh": 28298,
+ "Poly": 28299,
+ "▁cinq": 28300,
+ "▁Above": 28301,
+ "▁Joshua": 28302,
+ "ZERO": 28303,
+ "▁summit": 28304,
+ "▁Urs": 28305,
+ "▁curl": 28306,
+ "▁visa": 28307,
+ "▁resur": 28308,
+ "={'": 28309,
+ "feat": 28310,
+ "▁absorb": 28311,
+ "▁planets": 28312,
+ "▁princess": 28313,
+ "▁Jahrhunderts": 28314,
+ "xp": 28315,
+ "▁NBC": 28316,
+ "▁коми": 28317,
+ "▁FUN": 28318,
+ "▁neuen": 28319,
+ "▁déjà": 28320,
+ "▁Oz": 28321,
+ "bben": 28322,
+ "VIDEO": 28323,
+ "▁ejempl": 28324,
+ "▁considers": 28325,
+ "atri": 28326,
+ "▁arrog": 28327,
+ "ioso": 28328,
+ "▁hace": 28329,
+ "▁contacted": 28330,
+ "▁unple": 28331,
+ "▁sponsored": 28332,
+ "▁trainer": 28333,
+ "sbi": 28334,
+ "▁занима": 28335,
+ "Criterion": 28336,
+ "ното": 28337,
+ "scheme": 28338,
+ "ennial": 28339,
+ "perform": 28340,
+ "▁fixing": 28341,
+ "▁постро": 28342,
+ "arb": 28343,
+ "EXIT": 28344,
+ "▁café": 28345,
+ "ituted": 28346,
+ "riages": 28347,
+ "Tur": 28348,
+ "▁haber": 28349,
+ "elasticsearch": 28350,
+ "▁ал": 28351,
+ "rh": 28352,
+ "▁voll": 28353,
+ "CLU": 28354,
+ "Mil": 28355,
+ "▁membres": 28356,
+ "▁remarked": 28357,
+ "вана": 28358,
+ "=\"_": 28359,
+ "Less": 28360,
+ "(\"\");": 28361,
+ "▁Yale": 28362,
+ "berries": 28363,
+ "▁releasing": 28364,
+ "▁imports": 28365,
+ "idea": 28366,
+ "▁(+": 28367,
+ "▁arqu": 28368,
+ "ificación": 28369,
+ "▁пара": 28370,
+ "▁Rangers": 28371,
+ "Mic": 28372,
+ "▁nederbörd": 28373,
+ "▁imaginary": 28374,
+ "▁specialists": 28375,
+ "▁hoof": 28376,
+ "Modules": 28377,
+ "▁sadly": 28378,
+ "ceil": 28379,
+ "TabIndex": 28380,
+ "ationale": 28381,
+ "▁Partner": 28382,
+ "tbody": 28383,
+ "▁leverage": 28384,
+ "DN": 28385,
+ "▁Prec": 28386,
+ "▁Sé": 28387,
+ "▁Mam": 28388,
+ "▁afin": 28389,
+ "isValid": 28390,
+ "Pse": 28391,
+ "▁сторо": 28392,
+ "▁chopped": 28393,
+ "▁Minor": 28394,
+ "▁dabei": 28395,
+ "David": 28396,
+ "ussia": 28397,
+ "▁деревня": 28398,
+ "▁Identity": 28399,
+ "▁LGBT": 28400,
+ "ције": 28401,
+ "▁Orts": 28402,
+ "▁parti": 28403,
+ "▁Bachelor": 28404,
+ "uga": 28405,
+ "▁OPT": 28406,
+ "▁Seth": 28407,
+ "▁LIABLE": 28408,
+ "▁inaugur": 28409,
+ "▁Shanghai": 28410,
+ "▁relaxing": 28411,
+ "циона": 28412,
+ "\"%": 28413,
+ "▁obey": 28414,
+ "▁Airlines": 28415,
+ "Links": 28416,
+ "▁Celt": 28417,
+ "▁Admin": 28418,
+ "agation": 28419,
+ "▁worries": 28420,
+ "INTE": 28421,
+ "arith": 28422,
+ "Fatalf": 28423,
+ "]])": 28424,
+ "colm": 28425,
+ "▁archae": 28426,
+ "▁brushed": 28427,
+ "▁tät": 28428,
+ "▁structured": 28429,
+ "тии": 28430,
+ "▁homem": 28431,
+ "[:,": 28432,
+ "▁navy": 28433,
+ "getKey": 28434,
+ "powered": 28435,
+ "▁sucked": 28436,
+ "▁zomb": 28437,
+ "issant": 28438,
+ "▁Might": 28439,
+ "▁Pull": 28440,
+ "rir": 28441,
+ "▁пі": 28442,
+ "▁seas": 28443,
+ "▁Wrest": 28444,
+ "▁tense": 28445,
+ "▁atm": 28446,
+ "▁havet": 28447,
+ "▁pierws": 28448,
+ "▁tragic": 28449,
+ "▁Diff": 28450,
+ "▁confidential": 28451,
+ "successful": 28452,
+ "ęż": 28453,
+ "▁Chain": 28454,
+ "▁Kenya": 28455,
+ "Choice": 28456,
+ "ocur": 28457,
+ "aniu": 28458,
+ "▁consultant": 28459,
+ "▁Advis": 28460,
+ "Lif": 28461,
+ "▁Lors": 28462,
+ "avorite": 28463,
+ "▁utilizing": 28464,
+ "▁vintage": 28465,
+ "Matcher": 28466,
+ "▁membre": 28467,
+ "▁Expect": 28468,
+ "▁tracing": 28469,
+ "nog": 28470,
+ "▁dej": 28471,
+ "▁уче": 28472,
+ "▁loops": 28473,
+ "▁onclick": 28474,
+ "▁GPU": 28475,
+ "▁Albums": 28476,
+ "▁Archives": 28477,
+ "вата": 28478,
+ "▁stove": 28479,
+ "шли": 28480,
+ "ancies": 28481,
+ "▁gemeente": 28482,
+ "mob": 28483,
+ "PDF": 28484,
+ "eso": 28485,
+ "▁vég": 28486,
+ "Resolve": 28487,
+ "▁teaches": 28488,
+ "ложе": 28489,
+ "▁ство": 28490,
+ "▁Одна": 28491,
+ "▁fid": 28492,
+ "Something": 28493,
+ "▁nebo": 28494,
+ "▁Valentine": 28495,
+ "rowning": 28496,
+ "▁але": 28497,
+ "awi": 28498,
+ "ishi": 28499,
+ "▁SPI": 28500,
+ "▁spel": 28501,
+ "▁біль": 28502,
+ "▁participant": 28503,
+ "▁Ned": 28504,
+ "▁Gast": 28505,
+ "▁blond": 28506,
+ "▁saves": 28507,
+ "colored": 28508,
+ "▁ACTION": 28509,
+ "▁Politiker": 28510,
+ "}$)": 28511,
+ "▁Dum": 28512,
+ "dentry": 28513,
+ "Student": 28514,
+ "▁~=": 28515,
+ "loads": 28516,
+ "▁Foster": 28517,
+ "一个": 28518,
+ "▁PK": 28519,
+ "▁SB": 28520,
+ "▁Hern": 28521,
+ "▁Exhib": 28522,
+ "Listeners": 28523,
+ "Sun": 28524,
+ "plac": 28525,
+ "▁Bever": 28526,
+ "▁incluy": 28527,
+ "▁dc": 28528,
+ "argc": 28529,
+ "▁ged": 28530,
+ "спа": 28531,
+ "▁Formula": 28532,
+ "▁сем": 28533,
+ "▁empt": 28534,
+ "unregister": 28535,
+ "▁Queensland": 28536,
+ "ández": 28537,
+ "otive": 28538,
+ "▁alley": 28539,
+ "▁Democrat": 28540,
+ "▁travail": 28541,
+ "▁$,": 28542,
+ "RP": 28543,
+ "рое": 28544,
+ "personal": 28545,
+ "▁période": 28546,
+ "HOME": 28547,
+ "omes": 28548,
+ "▁recognised": 28549,
+ "heng": 28550,
+ "▁Jung": 28551,
+ "▁Roland": 28552,
+ "▁convicted": 28553,
+ "Locked": 28554,
+ "▁mari": 28555,
+ "▁Luxem": 28556,
+ "referto": 28557,
+ "Deleted": 28558,
+ "intent": 28559,
+ "▁Staats": 28560,
+ "▁області": 28561,
+ "ит": 28562,
+ "▁саве": 28563,
+ "▁Protocol": 28564,
+ "ając": 28565,
+ "chk": 28566,
+ "TypeInfo": 28567,
+ "▁pkt": 28568,
+ "▁scandal": 28569,
+ "▁individually": 28570,
+ "FMT": 28571,
+ "▁nj": 28572,
+ "abile": 28573,
+ "▁Rivers": 28574,
+ "PROPERTY": 28575,
+ "VB": 28576,
+ "wort": 28577,
+ "▁splitting": 28578,
+ "achten": 28579,
+ "▁ARISING": 28580,
+ "▁sip": 28581,
+ "▁fres": 28582,
+ "▁groom": 28583,
+ "Hol": 28584,
+ "▁canon": 28585,
+ "▁abruptly": 28586,
+ "▁afterward": 28587,
+ "▁Running": 28588,
+ "▁ji": 28589,
+ "▁%,": 28590,
+ "▁Palestinian": 28591,
+ "RW": 28592,
+ "pgfscope": 28593,
+ "▁countryside": 28594,
+ "▁fortunate": 28595,
+ "▁cél": 28596,
+ "▁Pointer": 28597,
+ "ensors": 28598,
+ "rating": 28599,
+ "▁buffers": 28600,
+ "▁remot": 28601,
+ "▁PropTypes": 28602,
+ "▁Nah": 28603,
+ "altern": 28604,
+ "▁easiest": 28605,
+ "▁invas": 28606,
+ "▁clk": 28607,
+ "copyright": 28608,
+ "▁blanc": 28609,
+ "SAMP": 28610,
+ "▁Cohen": 28611,
+ "▁Shell": 28612,
+ "▁destroying": 28613,
+ "▁Zel": 28614,
+ "dater": 28615,
+ "čen": 28616,
+ "▁filing": 28617,
+ "▁integrate": 28618,
+ "xit": 28619,
+ "▁RET": 28620,
+ "lene": 28621,
+ "calls": 28622,
+ "▁slaughter": 28623,
+ "initialized": 28624,
+ "unches": 28625,
+ "▁Trace": 28626,
+ "efficient": 28627,
+ "▁Woods": 28628,
+ "▁longitud": 28629,
+ "GN": 28630,
+ "▁Kont": 28631,
+ "▁chunks": 28632,
+ "ách": 28633,
+ "▁unemployment": 28634,
+ "acom": 28635,
+ "▁slowed": 28636,
+ "▁outlined": 28637,
+ "xffff": 28638,
+ "▁ikke": 28639,
+ "▁workspace": 28640,
+ "Mc": 28641,
+ "▁kicking": 28642,
+ "▁embedding": 28643,
+ "chnitt": 28644,
+ "erten": 28645,
+ "▁Interior": 28646,
+ "▁Songs": 28647,
+ "mmc": 28648,
+ "▁analyzed": 28649,
+ "▁Coupe": 28650,
+ "▁favorites": 28651,
+ "▁tt": 28652,
+ "▁той": 28653,
+ "Routing": 28654,
+ "▁Silva": 28655,
+ "▁anderem": 28656,
+ "▁honom": 28657,
+ "▁использова": 28658,
+ ".\"]": 28659,
+ "▁Wu": 28660,
+ "legt": 28661,
+ "▁spoon": 28662,
+ "▁jap": 28663,
+ "▁Extension": 28664,
+ "erne": 28665,
+ "▁vagy": 28666,
+ "▁села": 28667,
+ "▁функ": 28668,
+ "▁analytics": 28669,
+ "▁sug": 28670,
+ "▁Async": 28671,
+ "▁peaks": 28672,
+ "▁Gym": 28673,
+ "▁lawsuit": 28674,
+ "<>": 28675,
+ "ialis": 28676,
+ "etric": 28677,
+ "faced": 28678,
+ "▁disrupt": 28679,
+ "▁få": 28680,
+ "Inputs": 28681,
+ "`);": 28682,
+ "▁Mend": 28683,
+ "gon": 28684,
+ "▁\",\"": 28685,
+ "▁nerves": 28686,
+ "▁doubts": 28687,
+ "sap": 28688,
+ "▁sow": 28689,
+ ",\\,\\": 28690,
+ "▁BS": 28691,
+ "▁Glad": 28692,
+ "▁aster": 28693,
+ "œuvre": 28694,
+ "▁Bangl": 28695,
+ "▁iPad": 28696,
+ "useppe": 28697,
+ "▁conducting": 28698,
+ "▁({\\": 28699,
+ "▁Harbor": 28700,
+ "psz": 28701,
+ "▁FIFA": 28702,
+ "_**": 28703,
+ "emor": 28704,
+ "▁": 28705,
+ "e": 28706,
+ "t": 28707,
+ "a": 28708,
+ "o": 28709,
+ "i": 28710,
+ "n": 28711,
+ "r": 28712,
+ "s": 28713,
+ "l": 28714,
+ "d": 28715,
+ "h": 28716,
+ "c": 28717,
+ "u": 28718,
+ "m": 28719,
+ "p": 28720,
+ "g": 28721,
+ "f": 28722,
+ ".": 28723,
+ "y": 28724,
+ ",": 28725,
+ "b": 28726,
+ "w": 28727,
+ "v": 28728,
+ "k": 28729,
+ "_": 28730,
+ ")": 28731,
+ "(": 28732,
+ "-": 28733,
+ "0": 28734,
+ "S": 28735,
+ "*": 28736,
+ "I": 28737,
+ "T": 28738,
+ "\"": 28739,
+ "1": 28740,
+ "A": 28741,
+ "'": 28742,
+ "C": 28743,
+ "x": 28744,
+ ";": 28745,
+ "=": 28746,
+ ":": 28747,
+ "/": 28748,
+ "E": 28749,
+ "2": 28750,
+ "{": 28751,
+ "}": 28752,
+ "P": 28753,
+ "R": 28754,
+ "M": 28755,
+ "\\": 28756,
+ "D": 28757,
+ "L": 28758,
+ "N": 28759,
+ "B": 28760,
+ "о": 28761,
+ "O": 28762,
+ "а": 28763,
+ "z": 28764,
+ "F": 28765,
+ "|": 28766,
+ ">": 28767,
+ "j": 28768,
+ "H": 28769,
+ "3": 28770,
+ "#": 28771,
+ "и": 28772,
+ "е": 28773,
+ "9": 28774,
+ "q": 28775,
+ "$": 28776,
+ "G": 28777,
+ "н": 28778,
+ "U": 28779,
+ "W": 28780,
+ "4": 28781,
+ "5": 28782,
+ "8": 28783,
+ "6": 28784,
+ "р": 28785,
+ "т": 28786,
+ "7": 28787,
+ "с": 28788,
+ "<": 28789,
+ "V": 28790,
+ "в": 28791,
+ "[": 28792,
+ "]": 28793,
+ "л": 28794,
+ "к": 28795,
+ "K": 28796,
+ "é": 28797,
+ "J": 28798,
+ "д": 28799,
+ "&": 28800,
+ "\r": 28801,
+ "Y": 28802,
+ "м": 28803,
+ "?": 28804,
+ "у": 28805,
+ "+": 28806,
+ "п": 28807,
+ "!": 28808,
+ "’": 28809,
+ "г": 28810,
+ "я": 28811,
+ "з": 28812,
+ "і": 28813,
+ "X": 28814,
+ "^": 28815,
+ "–": 28816,
+ "б": 28817,
+ "@": 28818,
+ "й": 28819,
+ "á": 28820,
+ "—": 28821,
+ "ь": 28822,
+ "%": 28823,
+ "Q": 28824,
+ "ó": 28825,
+ "ч": 28826,
+ "í": 28827,
+ "Z": 28828,
+ "ы": 28829,
+ "ä": 28830,
+ "х": 28831,
+ "`": 28832,
+ "ц": 28833,
+ "ö": 28834,
+ "“": 28835,
+ "ж": 28836,
+ "ü": 28837,
+ "”": 28838,
+ "à": 28839,
+ "è": 28840,
+ "ш": 28841,
+ "ю": 28842,
+ "ł": 28843,
+ "С": 28844,
+ "~": 28845,
+ "ф": 28846,
+ "П": 28847,
+ "»": 28848,
+ "В": 28849,
+ "«": 28850,
+ "å": 28851,
+ "К": 28852,
+ "щ": 28853,
+ "·": 28854,
+ "ј": 28855,
+ "М": 28856,
+ "ç": 28857,
+ "А": 28858,
+ "Н": 28859,
+ "Р": 28860,
+ "Б": 28861,
+ "č": 28862,
+ "ú": 28863,
+ "ę": 28864,
+ "ã": 28865,
+ "ą": 28866,
+ "ă": 28867,
+ "Д": 28868,
+ "ї": 28869,
+ "ъ": 28870,
+ "ě": 28871,
+ "Г": 28872,
+ "š": 28873,
+ "О": 28874,
+ "Т": 28875,
+ "ê": 28876,
+ "ñ": 28877,
+ "…": 28878,
+ "ž": 28879,
+ "ß": 28880,
+ "ё": 28881,
+ "ż": 28882,
+ "ř": 28883,
+ "ś": 28884,
+ "Л": 28885,
+ "ő": 28886,
+ "„": 28887,
+ "э": 28888,
+ "ý": 28889,
+ "У": 28890,
+ "â": 28891,
+ "И": 28892,
+ "є": 28893,
+ "‘": 28894,
+ "î": 28895,
+ "З": 28896,
+ "Ф": 28897,
+ "ò": 28898,
+ "•": 28899,
+ "ć": 28900,
+ "É": 28901,
+ "°": 28902,
+ "ș": 28903,
+ "Х": 28904,
+ "ț": 28905,
+ "ô": 28906,
+ "Е": 28907,
+ "ń": 28908,
+ "Ч": 28909,
+ "Ш": 28910,
+ "ø": 28911,
+ "ù": 28912,
+ "ů": 28913,
+ "的": 28914,
+ "ا": 28915,
+ "æ": 28916,
+ "њ": 28917,
+ "љ": 28918,
+ "ë": 28919,
+ "ï": 28920,
+ "Э": 28921,
+ "£": 28922,
+ "−": 28923,
+ ",": 28924,
+ "õ": 28925,
+ "ћ": 28926,
+ "": 28927,
+ "Ц": 28928,
+ "І": 28929,
+ "ā": 28930,
+ "ű": 28931,
+ "†": 28932,
+ "ل": 28933,
+ "ō": 28934,
+ "": 28935,
+ "º": 28936,
+ "Я": 28937,
+ "′": 28938,
+ "Á": 28939,
+ "Ö": 28940,
+ "²": 28941,
+ "Ж": 28942,
+ "ì": 28943,
+ "。": 28944,
+ "数": 28945,
+ "×": 28946,
+ "ر": 28947,
+ "α": 28948,
+ "́": 28949,
+ "Ю": 28950,
+ "û": 28951,
+ "œ": 28952,
+ "ı": 28953,
+ "م": 28954,
+ "ن": 28955,
+ "ª": 28956,
+ "ź": 28957,
+ "ο": 28958,
+ "″": 28959,
+ "€": 28960,
+ "Ü": 28961,
+ "و": 28962,
+ "用": 28963,
+ "À": 28964,
+ "Č": 28965,
+ "Š": 28966,
+ "ت": 28967,
+ "د": 28968,
+ "一": 28969,
+ "¿": 28970,
+ "是": 28971,
+ "ي": 28972,
+ "ђ": 28973,
+ "®": 28974,
+ "ی": 28975,
+ "ν": 28976,
+ "đ": 28977,
+ "τ": 28978,
+ "─": 28979,
+ "ι": 28980,
+ "ε": 28981,
+ "→": 28982,
+ "ب": 28983,
+ "Å": 28984,
+ "ū": 28985,
+ "№": 28986,
+ "ş": 28987,
+ "不": 28988,
+ "џ": 28989,
+ "ー": 28990,
+ "中": 28991,
+ "Î": 28992,
+ "の": 28993,
+ ":": 28994,
+ "个": 28995,
+ "Й": 28996,
+ "ρ": 28997,
+ "有": 28998,
+ "Ä": 28999,
+ " ": 29000,
+ "ī": 29001,
+ "©": 29002,
+ "为": 29003,
+ "ه": 29004,
+ "י": 29005,
+ "ו": 29006,
+ "时": 29007,
+ "س": 29008,
+ "Ś": 29009,
+ "在": 29010,
+ "件": 29011,
+ "取": 29012,
+ "ς": 29013,
+ "™": 29014,
+ "이": 29015,
+ "σ": 29016,
+ "μ": 29017,
+ "定": 29018,
+ "文": 29019,
+ "据": 29020,
+ "置": 29021,
+ "Ž": 29022,
+ "±": 29023,
+ "表": 29024,
+ "成": 29025,
+ "ň": 29026,
+ "λ": 29027,
+ "¡": 29028,
+ "È": 29029,
+ "π": 29030,
+ "字": 29031,
+ "│": 29032,
+ "Ј": 29033,
+ "回": 29034,
+ "Є": 29035,
+ "到": 29036,
+ "行": 29037,
+ "§": 29038,
+ "½": 29039,
+ "ع": 29040,
+ "、": 29041,
+ "Ł": 29042,
+ "다": 29043,
+ "ン": 29044,
+ "κ": 29045,
+ "名": 29046,
+ "ה": 29047,
+ "入": 29048,
+ "η": 29049,
+ "大": 29050,
+ "对": 29051,
+ "可": 29052,
+ "Â": 29053,
+ "上": 29054,
+ "█": 29055,
+ "新": 29056,
+ "ف": 29057,
+ "加": 29058,
+ "要": 29059,
+ "Ż": 29060,
+ "下": 29061,
+ "分": 29062,
+ "值": 29063,
+ "ת": 29064,
+ "出": 29065,
+ "类": 29066,
+ "请": 29067,
+ "": 29068,
+ "息": 29069,
+ "Ú": 29070,
+ "υ": 29071,
+ "获": 29072,
+ "示": 29073,
+ "以": 29074,
+ "ר": 29075,
+ "接": 29076,
+ "ל": 29077,
+ "を": 29078,
+ "存": 29079,
+ "信": 29080,
+ "设": 29081,
+ "方": 29082,
+ "ش": 29083,
+ "能": 29084,
+ "点": 29085,
+ "人": 29086,
+ "前": 29087,
+ "ğ": 29088,
+ "作": 29089,
+ "═": 29090,
+ "↘": 29091,
+ "ð": 29092,
+ "理": 29093,
+ "■": 29094,
+ "法": 29095,
+ "️": 29096,
+ "ˈ": 29097,
+ "果": 29098,
+ "发": 29099,
+ "ح": 29100,
+ "γ": 29101,
+ "ɵ": 29102,
+ "า": 29103,
+ "َ": 29104,
+ "了": 29105,
+ "户": 29106,
+ "Í": 29107,
+ "ə": 29108,
+ "ス": 29109,
+ "查": 29110,
+ "し": 29111,
+ "מ": 29112,
+ "单": 29113,
+ "ť": 29114,
+ "ق": 29115,
+ "る": 29116,
+ "间": 29117,
+ "如": 29118,
+ "本": 29119,
+ "后": 29120,
+ "ί": 29121,
+ "式": 29122,
+ "ト": 29123,
+ "Щ": 29124,
+ "Ó": 29125,
+ "す": 29126,
+ "א": 29127,
+ "生": 29128,
+ "动": 29129,
+ "ک": 29130,
+ "和": 29131,
+ "い": 29132,
+ "": 29133,
+ "ა": 29134,
+ "가": 29135,
+ "하": 29136,
+ "�": 29137,
+ "小": 29138,
+ "返": 29139,
+ "否": 29140,
+ "ة": 29141,
+ "日": 29142,
+ "로": 29143,
+ "标": 29144,
+ "码": 29145,
+ "地": 29146,
+ "位": 29147,
+ "에": 29148,
+ " ": 29149,
+ "列": 29150,
+ "수": 29151,
+ "β": 29152,
+ "除": 29153,
+ "使": 29154,
+ "ש": 29155,
+ "ج": 29156,
+ "イ": 29157,
+ "δ": 29158,
+ "自": 29159,
+ "于": 29160,
+ "지": 29161,
+ "当": 29162,
+ "所": 29163,
+ "기": 29164,
+ "ი": 29165,
+ "ב": 29166,
+ "ร": 29167,
+ "★": 29168,
+ "子": 29169,
+ "号": 29170,
+ "ك": 29171,
+ "参": 29172,
+ "型": 29173,
+ "に": 29174,
+ "는": 29175,
+ "这": 29176,
+ "开": 29177,
+ "น": 29178,
+ "会": 29179,
+ "器": 29180,
+ "面": 29181,
+ "ル": 29182,
+ "图": 29183,
+ "度": 29184,
+ ")": 29185,
+ "(": 29186,
+ "의": 29187,
+ "内": 29188,
+ "을": 29189,
+ "最": 29190,
+ "": 29191,
+ "化": 29192,
+ "建": 29193,
+ "니": 29194,
+ "量": 29195,
+ "😂": 29196,
+ "始": 29197,
+ "ē": 29198,
+ "خ": 29199,
+ "를": 29200,
+ "ά": 29201,
+ "过": 29202,
+ "³": 29203,
+ "´": 29204,
+ "组": 29205,
+ "功": 29206,
+ "": 29207,
+ "": 29208,
+ "区": 29209,
+ "ز": 29210,
+ "ґ": 29211,
+ "ό": 29212,
+ "ッ": 29213,
+ "ω": 29214,
+ "Ç": 29215,
+ "选": 29216,
+ "通": 29217,
+ "结": 29218,
+ "录": 29219,
+ "改": 29220,
+ "ク": 29221,
+ "目": 29222,
+ "指": 29223,
+ "务": 29224,
+ "๐": 29225,
+ "输": 29226,
+ "た": 29227,
+ "อ": 29228,
+ "关": 29229,
+ "で": 29230,
+ "调": 29231,
+ "ा": 29232,
+ "정": 29233,
+ "合": 29234,
+ "已": 29235,
+ "시": 29236,
+ "部": 29237,
+ "页": 29238,
+ "━": 29239,
+ "ː": 29240,
+ "ま": 29241,
+ "我": 29242,
+ "求": 29243,
+ "市": 29244,
+ "次": 29245,
+ "נ": 29246,
+ "实": 29247,
+ "将": 29248,
+ "重": 29249,
+ "更": 29250,
+ "制": 29251,
+ "符": 29252,
+ "配": 29253,
+ "象": 29254,
+ "θ": 29255,
+ "ก": 29256,
+ "て": 29257,
+ "进": 29258,
+ "需": 29259,
+ "Đ": 29260,
+ "性": 29261,
+ "认": 29262,
+ "来": 29263,
+ "题": 29264,
+ "程": 29265,
+ "模": 29266,
+ "!": 29267,
+ "失": 29268,
+ "口": 29269,
+ "な": 29270,
+ "έ": 29271,
+ "": 29272,
+ "空": 29273,
+ "": 29274,
+ "期": 29275,
+ "者": 29276,
+ "は": 29277,
+ "Ђ": 29278,
+ "提": 29279,
+ "ή": 29280,
+ "ラ": 29281,
+ "한": 29282,
+ "态": 29283,
+ "复": 29284,
+ "ง": 29285,
+ "ე": 29286,
+ "Ø": 29287,
+ "리": 29288,
+ "修": 29289,
+ "‚": 29290,
+ "得": 29291,
+ "多": 29292,
+ "格": 29293,
+ "자": 29294,
+ "ע": 29295,
+ "่": 29296,
+ "函": 29297,
+ "应": 29298,
+ "↗": 29299,
+ "्": 29300,
+ "เ": 29301,
+ "正": 29302,
+ "注": 29303,
+ "스": 29304,
+ "서": 29305,
+ "リ": 29306,
+ "φ": 29307,
+ "ص": 29308,
+ "が": 29309,
+ "则": 29310,
+ "消": 29311,
+ "节": 29312,
+ "序": 29313,
+ "代": 29314,
+ "사": 29315,
+ "と": 29316,
+ "ד": 29317,
+ "้": 29318,
+ "र": 29319,
+ "此": 29320,
+ "保": 29321,
+ "ア": 29322,
+ "ư": 29323,
+ "인": 29324,
+ "ė": 29325,
+ "处": 29326,
+ "删": 29327,
+ "ɛ": 29328,
+ "容": 29329,
+ "ط": 29330,
+ "": 29331,
+ "之": 29332,
+ "包": 29333,
+ "状": 29334,
+ "ド": 29335,
+ "İ": 29336,
+ "体": 29337,
+ "同": 29338,
+ "事": 29339,
+ "🙂": 29340,
+ "タ": 29341,
+ "χ": 29342,
+ "ʿ": 29343,
+ "Ș": 29344,
+ "主": 29345,
+ "品": 29346,
+ "ק": 29347,
+ "询": 29348,
+ "创": 29349,
+ "该": 29350,
+ " ": 29351,
+ "元": 29352,
+ "第": 29353,
+ "天": 29354,
+ "或": 29355,
+ "年": 29356,
+ "转": 29357,
+ "ח": 29358,
+ "传": 29359,
+ "ţ": 29360,
+ "路": 29361,
+ "例": 29362,
+ "机": 29363,
+ "Ã": 29364,
+ "ď": 29365,
+ "高": 29366,
+ "相": 29367,
+ "โ": 29368,
+ "片": 29369,
+ "―": 29370,
+ "操": 29371,
+ "ա": 29372,
+ "ม": 29373,
+ "全": 29374,
+ "无": 29375,
+ "月": 29376,
+ "称": 29377,
+ "ั": 29378,
+ "就": 29379,
+ "": 29380,
+ "明": 29381,
+ "计": 29382,
+ "你": 29383,
+ "败": 29384,
+ "密": 29385,
+ "解": 29386,
+ "れ": 29387,
+ "أ": 29388,
+ "变": 29389,
+ "段": 29390,
+ "条": 29391,
+ "默": 29392,
+ "●": 29393,
+ "ล": 29394,
+ "色": 29395,
+ "断": 29396,
+ "商": 29397,
+ "ם": 29398,
+ "か": 29399,
+ "里": 29400,
+ "系": 29401,
+ "编": 29402,
+ "错": 29403,
+ "트": 29404,
+ "只": 29405,
+ "县": 29406,
+ "ს": 29407,
+ "常": 29408,
+ "初": 29409,
+ "ɔ": 29410,
+ "Α": 29411,
+ "フ": 29412,
+ "►": 29413,
+ "等": 29414,
+ "일": 29415,
+ "・": 29416,
+ "Ō": 29417,
+ "情": 29418,
+ "现": 29419,
+ "Ř": 29420,
+ "ِ": 29421,
+ "さ": 29422,
+ "ạ": 29423,
+ "용": 29424,
+ "证": 29425,
+ "해": 29426,
+ "手": 29427,
+ "支": 29428,
+ "입": 29429,
+ "服": 29430,
+ "்": 29431,
+ "道": 29432,
+ "어": 29433,
+ "送": 29434,
+ "载": 29435,
+ "限": 29436,
+ "线": 29437,
+ "属": 29438,
+ "": 29439,
+ "他": 29440,
+ "放": 29441,
+ "记": 29442,
+ "公": 29443,
+ "没": 29444,
+ "添": 29445,
+ "显": 29446,
+ "บ": 29447,
+ "ย": 29448,
+ "რ": 29449,
+ "其": 29450,
+ "集": 29451,
+ "金": 29452,
+ "国": 29453,
+ "任": 29454,
+ "ە": 29455,
+ "话": 29456,
+ "并": 29457,
+ "被": 29458,
+ "ύ": 29459,
+ "都": 29460,
+ "گ": 29461,
+ "意": 29462,
+ "כ": 29463,
+ "经": 29464,
+ "성": 29465,
+ "看": 29466,
+ "פ": 29467,
+ "址": 29468,
+ "ס": 29469,
+ "드": 29470,
+ "交": 29471,
+ "¼": 29472,
+ "Џ": 29473,
+ "完": 29474,
+ "Δ": 29475,
+ "义": 29476,
+ "보": 29477,
+ "向": 29478,
+ "换": 29479,
+ "山": 29480,
+ "算": 29481,
+ "二": 29482,
+ "پ": 29483,
+ "⁄": 29484,
+ "判": 29485,
+ "级": 29486,
+ "工": 29487,
+ "ด": 29488,
+ "⠀": 29489,
+ "家": 29490,
+ "レ": 29491,
+ "三": 29492,
+ "原": 29493,
+ "】": 29494,
+ "长": 29495,
+ "া": 29496,
+ "管": 29497,
+ "ѝ": 29498,
+ "क": 29499,
+ "学": 29500,
+ "ロ": 29501,
+ "验": 29502,
+ "写": 29503,
+ "Œ": 29504,
+ "从": 29505,
+ "【": 29506,
+ "收": 29507,
+ "ả": 29508,
+ "未": 29509,
+ "登": 29510,
+ "고": 29511,
+ "源": 29512,
+ "每": 29513,
+ "µ": 29514,
+ "误": 29515,
+ "り": 29516,
+ "요": 29517,
+ "按": 29518,
+ "ว": 29519,
+ "权": 29520,
+ "根": 29521,
+ "プ": 29522,
+ "串": 29523,
+ "ส": 29524,
+ "›": 29525,
+ "제": 29526,
+ "シ": 29527,
+ "Ş": 29528,
+ "确": 29529,
+ "好": 29530,
+ "统": 29531,
+ "效": 29532,
+ "网": 29533,
+ "\u0001": 29534,
+ "物": 29535,
+ "아": 29536,
+ "也": 29537,
+ "은": 29538,
+ "ệ": 29539,
+ "न": 29540,
+ "项": 29541,
+ "资": 29542,
+ "こ": 29543,
+ "引": 29544,
+ "ジ": 29545,
+ "ค": 29546,
+ "版": 29547,
+ "ท": 29548,
+ "平": 29549,
+ "们": 29550,
+ "与": 29551,
+ "き": 29552,
+ "移": 29553,
+ "ि": 29554,
+ "素": 29555,
+ "执": 29556,
+ "주": 29557,
+ "‐": 29558,
+ "Ґ": 29559,
+ "ี": 29560,
+ "板": 29561,
+ "问": 29562,
+ "Ε": 29563,
+ "安": 29564,
+ "면": 29565,
+ "소": 29566,
+ "ต": 29567,
+ "ิ": 29568,
+ "持": 29569,
+ "습": 29570,
+ "Σ": 29571,
+ "ら": 29572,
+ "コ": 29573,
+ "心": 29574,
+ "Π": 29575,
+ "打": 29576,
+ "」": 29577,
+ "상": 29578,
+ "「": 29579,
+ "检": 29580,
+ "库": 29581,
+ "÷": 29582,
+ "으": 29583,
+ "测": 29584,
+ "ん": 29585,
+ "े": 29586,
+ "ُ": 29587,
+ "力": 29588,
+ "直": 29589,
+ "由": 29590,
+ "ى": 29591,
+ "试": 29592,
+ "必": 29593,
+ "端": 29594,
+ "ʻ": 29595,
+ "先": 29596,
+ "↑": 29597,
+ "命": 29598,
+ "도": 29599,
+ "전": 29600,
+ "ห": 29601,
+ "员": 29602,
+ "ɪ": 29603,
+ "있": 29604,
+ "比": 29605,
+ "ṣ": 29606,
+ "時": 29607,
+ "择": 29608,
+ "ذ": 29609,
+ "テ": 29610,
+ "": 29611,
+ "构": 29612,
+ "备": 29613,
+ "그": 29614,
+ "链": 29615,
+ "说": 29616,
+ "ლ": 29617,
+ "ן": 29618,
+ "签": 29619,
+ "う": 29620,
+ "غ": 29621,
+ "ế": 29622,
+ "ض": 29623,
+ "ḥ": 29624,
+ "启": 29625,
+ "력": 29626,
+ "ო": 29627,
+ "付": 29628,
+ "მ": 29629,
+ "索": 29630,
+ "特": 29631,
+ "ג": 29632,
+ "西": 29633,
+ "대": 29634,
+ "├": 29635,
+ "": 29636,
+ "": 29637,
+ "外": 29638,
+ "צ": 29639,
+ "头": 29640,
+ "连": 29641,
+ "流": 29642,
+ "◄": 29643,
+ "デ": 29644,
+ "カ": 29645,
+ "র": 29646,
+ "오": 29647,
+ "找": 29648,
+ "清": 29649,
+ "🤣": 29650,
+ "去": 29651,
+ "₹": 29652,
+ "경": 29653,
+ "グ": 29654,
+ "ْ": 29655,
+ "¢": 29656,
+ "因": 29657,
+ "": 29658,
+ "Κ": 29659,
+ "增": 29660,
+ "知": 29661,
+ "¶": 29662,
+ "像": 29663,
+ "♥": 29664,
+ "터": 29665,
+ "く": 29666,
+ "ậ": 29667,
+ "メ": 29668,
+ "Æ": 29669,
+ "省": 29670,
+ "स": 29671,
+ "म": 29672,
+ "❤": 29673,
+ "あ": 29674,
+ "样": 29675,
+ "起": 29676,
+ "台": 29677,
+ "读": 29678,
+ "角": 29679,
+ "南": 29680,
+ "整": 29681,
+ "订": 29682,
+ "\f": 29683,
+ "ט": 29684,
+ "マ": 29685,
+ "্": 29686,
+ "우": 29687,
+ "ն": 29688,
+ "您": 29689,
+ "ئ": 29690,
+ "基": 29691,
+ "水": 29692,
+ "생": 29693,
+ "‑": 29694,
+ "나": 29695,
+ "画": 29696,
+ "描": 29697,
+ "击": 29698,
+ "っ": 29699,
+ "라": 29700,
+ "ნ": 29701,
+ "ր": 29702,
+ "业": 29703,
+ "ბ": 29704,
+ "别": 29705,
+ "♦": 29706,
+ "ィ": 29707,
+ "त": 29708,
+ "给": 29709,
+ "문": 29710,
+ "形": 29711,
+ "控": 29712,
+ "然": 29713,
+ "동": 29714,
+ "Њ": 29715,
+ "": 29716,
+ "东": 29717,
+ "ป": 29718,
+ "州": 29719,
+ "排": 29720,
+ "세": 29721,
+ "装": 29722,
+ "할": 29723,
+ "Ć": 29724,
+ "∞": 29725,
+ "海": 29726,
+ "城": 29727,
+ "键": 29728,
+ "径": 29729,
+ "호": 29730,
+ "화": 29731,
+ "្": 29732,
+ "料": 29733,
+ "ơ": 29734,
+ "ी": 29735,
+ "ウ": 29736,
+ "具": 29737,
+ "ブ": 29738,
+ "块": 29739,
+ "再": 29740,
+ "ố": 29741,
+ "电": 29742,
+ ";": 29743,
+ "위": 29744,
+ "两": 29745,
+ "而": 29746,
+ "장": 29747,
+ "آ": 29748,
+ "Ț": 29749,
+ "バ": 29750,
+ "还": 29751,
+ "令": 29752,
+ "キ": 29753,
+ "ّ": 29754,
+ "값": 29755,
+ "번": 29756,
+ "만": 29757,
+ "总": 29758,
+ "ल": 29759,
+ "▲": 29760,
+ "异": 29761,
+ "光": 29762,
+ "客": 29763,
+ "非": 29764,
+ "ị": 29765,
+ "": 29766,
+ "þ": 29767,
+ "設": 29768,
+ "述": 29769,
+ "합": 29770,
+ "?": 29771,
+ "✔": 29772,
+ "导": 29773,
+ "ṇ": 29774,
+ "부": 29775,
+ "˙": 29776,
+ "Τ": 29777,
+ "も": 29778,
+ "구": 29779,
+ "镇": 29780,
+ "작": 29781,
+ "░": 29782,
+ "步": 29783,
+ "ộ": 29784,
+ "活": 29785,
+ "พ": 29786,
+ "←": 29787,
+ "ǎ": 29788,
+ "จ": 29789,
+ "束": 29790,
+ "ـ": 29791,
+ "": 29792,
+ "那": 29793,
+ "प": 29794,
+ "エ": 29795,
+ "志": 29796,
+ "么": 29797,
+ "运": 29798,
+ "北": 29799,
+ "超": 29800,
+ "་": 29801,
+ "布": 29802,
+ "ώ": 29803,
+ "͡": 29804,
+ "少": 29805,
+ "파": 29806,
+ "ʃ": 29807,
+ "ム": 29808,
+ "": 29809,
+ "卡": 29810,
+ "ন": 29811,
+ "Μ": 29812,
+ "ɑ": 29813,
+ "😉": 29814,
+ "辑": 29815,
+ "원": 29816,
+ "美": 29817,
+ "产": 29818,
+ "利": 29819,
+ "모": 29820,
+ "联": 29821,
+ "界": 29822,
+ "체": 29823,
+ "种": 29824,
+ "王": 29825,
+ "ľ": 29826,
+ "여": 29827,
+ "메": 29828,
+ "域": 29829,
+ "ვ": 29830,
+ "立": 29831,
+ "록": 29832,
+ "게": 29833,
+ "إ": 29834,
+ "ṭ": 29835,
+ "神": 29836,
+ "ո": 29837,
+ "音": 29838,
+ "☆": 29839,
+ "Ñ": 29840,
+ "조": 29841,
+ "動": 29842,
+ "缓": 29843,
+ "과": 29844,
+ "报": 29845,
+ "ʼ": 29846,
+ "ា": 29847,
+ "되": 29848,
+ "ե": 29849,
+ "视": 29850,
+ "ช": 29851,
+ "详": 29852,
+ "แ": 29853,
+ "¦": 29854,
+ "把": 29855,
+ "க": 29856,
+ "ি": 29857,
+ "출": 29858,
+ "비": 29859,
+ "边": 29860,
+ "框": 29861,
+ "व": 29862,
+ "サ": 29863,
+ "Ι": 29864,
+ "Ο": 29865,
+ "オ": 29866,
+ "¾": 29867,
+ "历": 29868,
+ "ŏ": 29869,
+ "门": 29870,
+ "ข": 29871,
+ "含": 29872,
+ "¬": 29873,
+ "周": 29874,
+ "填": 29875,
+ "待": 29876,
+ "ะ": 29877,
+ "დ": 29878,
+ "Ї": 29879,
+ "额": 29880,
+ "음": 29881,
+ "四": 29882,
+ "だ": 29883,
+ "회": 29884,
+ "止": 29885,
+ "率": 29886,
+ "环": 29887,
+ "パ": 29888,
+ "래": 29889,
+ "闭": 29890,
+ "̀": 29891,
+ "语": 29892,
+ "개": 29893,
+ "身": 29894,
+ "藏": 29895,
+ "य": 29896,
+ "된": 29897,
+ "即": 29898,
+ "拉": 29899,
+ "선": 29900,
+ "변": 29901,
+ "≥": 29902,
+ "ุ": 29903,
+ "些": 29904,
+ "🤷": 29905,
+ "せ": 29906,
+ "左": 29907,
+ "ợ": 29908,
+ "右": 29909,
+ "ể": 29910,
+ "내": 29911,
+ "ּ": 29912,
+ "ז": 29913,
+ "ে": 29914,
+ "告": 29915,
+ "ấ": 29916,
+ "白": 29917,
+ "账": 29918,
+ "费": 29919,
+ "江": 29920,
+ "み": 29921,
+ "‹": 29922,
+ "์": 29923,
+ "": 29924,
+ "造": 29925,
+ "但": 29926,
+ "十": 29927,
+ "它": 29928,
+ "ं": 29929,
+ "ŋ": 29930,
+ "ў": 29931,
+ "セ": 29932,
+ "女": 29933,
+ "⣿": 29934,
+ "ի": 29935,
+ "京": 29936,
+ "触": 29937,
+ "함": 29938,
+ "들": 29939,
+ "Ā": 29940,
+ "": 29941,
+ "石": 29942,
+ "よ": 29943,
+ "田": 29944,
+ "易": 29945,
+ "规": 29946,
+ "展": 29947,
+ "¯": 29948,
+ "做": 29949,
+ "星": 29950,
+ "უ": 29951,
+ "✓": 29952,
+ "თ": 29953,
+ "供": 29954,
+ "명": 29955,
+ "ξ": 29956,
+ "己": 29957,
+ "且": 29958,
+ "插": 29959,
+ "景": 29960,
+ "切": 29961,
+ "ไ": 29962,
+ "없": 29963,
+ "ョ": 29964,
+ "及": 29965,
+ "Ν": 29966,
+ "미": 29967,
+ "ث": 29968,
+ "데": 29969,
+ "价": 29970,
+ "乡": 29971,
+ "ह": 29972,
+ "チ": 29973,
+ "真": 29974,
+ "太": 29975,
+ "ู": 29976,
+ "ダ": 29977,
+ "局": 29978,
+ "♂": 29979,
+ "退": 29980,
+ "ு": 29981,
+ "ক": 29982,
+ "ி": 29983,
+ "何": 29984,
+ "😭": 29985,
+ "¥": 29986,
+ "": 29987,
+ "≈": 29988,
+ "司": 29989,
+ "层": 29990,
+ "실": 29991,
+ "站": 29992,
+ "首": 29993,
+ "款": 29994,
+ "រ": 29995,
+ "間": 29996,
+ "ָ": 29997,
+ "저": 29998,
+ "监": 29999,
+ "ァ": 30000,
+ "册": 30001,
+ "案": 30002,
+ "ो": 30003,
+ "反": 30004,
+ "听": 30005,
+ "族": 30006,
+ "析": 30007,
+ "ื": 30008,
+ "秒": 30009,
+ "공": 30010,
+ "": 30011,
+ "🚀": 30012,
+ "거": 30013,
+ "재": 30014,
+ "": 30015,
+ "場": 30016,
+ "广": 30017,
+ "播": 30018,
+ "║": 30019,
+ "⋅": 30020,
+ "技": 30021,
+ "贴": 30022,
+ "想": 30023,
+ "ʁ": 30024,
+ "ớ": 30025,
+ "ャ": 30026,
+ "중": 30027,
+ "》": 30028,
+ "速": 30029,
+ "频": 30030,
+ "队": 30031,
+ "ำ": 30032,
+ "け": 30033,
+ "ु": 30034,
+ "≤": 30035,
+ "↓": 30036,
+ "须": 30037,
+ "菜": 30038,
+ "̃": 30039,
+ "剪": 30040,
+ "버": 30041,
+ "ェ": 30042,
+ "Λ": 30043,
+ "细": 30044,
+ "選": 30045,
+ "द": 30046,
+ "¹": 30047,
+ "许": 30048,
+ "ầ": 30049,
+ "世": 30050,
+ "ュ": 30051,
+ "ء": 30052,
+ "‡": 30053,
+ "候": 30054,
+ "共": 30055,
+ "크": 30056,
+ "ธ": 30057,
+ "설": 30058,
+ "快": 30059,
+ "友": 30060,
+ "ְ": 30061,
+ "车": 30062,
+ "推": 30063,
+ "花": 30064,
+ "言": 30065,
+ "چ": 30066,
+ "至": 30067,
+ "開": 30068,
+ "校": 30069,
+ "個": 30070,
+ "村": 30071,
+ "つ": 30072,
+ "▌": 30073,
+ "ப": 30074,
+ "결": 30075,
+ "ņ": 30076,
+ "优": 30077,
+ "ន": 30078,
+ "达": 30079,
+ "核": 30080,
+ "ナ": 30081,
+ "场": 30082,
+ "影": 30083,
+ "🏻": 30084,
+ "钮": 30085,
+ "ظ": 30086,
+ "Þ": 30087,
+ "▼": 30088,
+ "お": 30089,
+ "份": 30090,
+ "微": 30091,
+ "ờ": 30092,
+ "识": 30093,
+ "행": 30094,
+ "《": 30095,
+ "ใ": 30096,
+ "ọ": 30097,
+ "预": 30098,
+ "ব": 30099,
+ "த": 30100,
+ "": 30101,
+ "ų": 30102,
+ "마": 30103,
+ "않": 30104,
+ "ɡ": 30105,
+ "계": 30106,
+ "연": 30107,
+ "五": 30108,
+ "Ź": 30109,
+ "め": 30110,
+ "很": 30111,
+ "간": 30112,
+ "無": 30113,
+ "ប": 30114,
+ "社": 30115,
+ "Ê": 30116,
+ "书": 30117,
+ "顶": 30118,
+ "ტ": 30119,
+ "才": 30120,
+ "云": 30121,
+ "└": 30122,
+ "ζ": 30123,
+ "،": 30124,
+ "搜": 30125,
+ "신": 30126,
+ "유": 30127,
+ "": 30128,
+ "✅": 30129,
+ "⭐": 30130,
+ "照": 30131,
+ "短": 30132,
+ "川": 30133,
+ "後": 30134,
+ "范": 30135,
+ "民": 30136,
+ "治": 30137,
+ "章": 30138,
+ "ề": 30139,
+ "바": 30140,
+ "ә": 30141,
+ "⚭": 30142,
+ "河": 30143,
+ "论": 30144,
+ "え": 30145,
+ "Ω": 30146,
+ "√": 30147,
+ "Ă": 30148,
+ "Γ": 30149,
+ "坐": 30150,
+ "적": 30151,
+ "停": 30152,
+ "추": 30153,
+ "受": 30154,
+ "♀": 30155,
+ "ʾ": 30156,
+ "树": 30157,
+ "林": 30158,
+ "치": 30159,
+ "fi": 30160,
+ "▒": 30161,
+ "张": 30162,
+ "着": 30163,
+ "访": 30164,
+ "考": 30165,
+ "教": 30166,
+ "ग": 30167,
+ "准": 30168,
+ "印": 30169,
+ "精": 30170,
+ "窗": 30171,
+ "宝": 30172,
+ "ち": 30173,
+ "围": 30174,
+ "ַ": 30175,
+ "致": 30176,
+ "モ": 30177,
+ "때": 30178,
+ "随": 30179,
+ "储": 30180,
+ "况": 30181,
+ "邮": 30182,
+ "武": 30183,
+ "⛔": 30184,
+ "维": 30185,
+ "ү": 30186,
+ "跳": 30187,
+ "ब": 30188,
+ "投": 30189,
+ "ủ": 30190,
+ "표": 30191,
+ "반": 30192,
+ "英": 30193,
+ "ʰ": 30194,
+ "👍": 30195,
+ "ज": 30196,
+ "带": 30197,
+ "為": 30198,
+ "续": 30199,
+ "ɨ": 30200,
+ "처": 30201,
+ "₂": 30202,
+ "클": 30203,
+ "群": 30204,
+ "현": 30205,
+ "风": 30206,
+ "购": 30207,
+ "ក": 30208,
+ "老": 30209,
+ "留": 30210,
+ "球": 30211,
+ "프": 30212,
+ "▄": 30213,
+ "史": 30214,
+ "Љ": 30215,
+ "⟩": 30216,
+ "분": 30217,
+ "გ": 30218,
+ "店": 30219,
+ "审": 30220,
+ "료": 30221,
+ "목": 30222,
+ "略": 30223,
+ "관": 30224,
+ "ִ": 30225,
+ "科": 30226,
+ "货": 30227,
+ "ம": 30228,
+ "络": 30229,
+ "阳": 30230,
+ "Ḥ": 30231,
+ "資": 30232,
+ "若": 30233,
+ "স": 30234,
+ "ہ": 30235,
+ "宽": 30236,
+ "见": 30237,
+ "ズ": 30238,
+ "游": 30239,
+ "방": 30240,
+ "ồ": 30241,
+ "ɾ": 30242,
+ "열": 30243,
+ "러": 30244,
+ "ך": 30245,
+ "\u001b": 30246,
+ "်": 30247,
+ "余": 30248,
+ "响": 30249,
+ "缩": 30250,
+ "ட": 30251,
+ "评": 30252,
+ "允": 30253,
+ "离": 30254,
+ "🤔": 30255,
+ "Ё": 30256,
+ "ʊ": 30257,
+ "黑": 30258,
+ "马": 30259,
+ "⟨": 30260,
+ "値": 30261,
+ "箱": 30262,
+ "야": 30263,
+ "ម": 30264,
+ "Ő": 30265,
+ "感": 30266,
+ "ツ": 30267,
+ "ụ": 30268,
+ "ポ": 30269,
+ "확": 30270,
+ "声": 30271,
+ "战": 30272,
+ "ѕ": 30273,
+ "変": 30274,
+ "와": 30275,
+ "父": 30276,
+ "ベ": 30277,
+ "助": 30278,
+ "업": 30279,
+ "ʲ": 30280,
+ "ÿ": 30281,
+ "充": 30282,
+ "强": 30283,
+ "博": 30284,
+ "ミ": 30285,
+ "销": 30286,
+ "당": 30287,
+ "記": 30288,
+ "什": 30289,
+ "匹": 30290,
+ "ւ": 30291,
+ "そ": 30292,
+ "코": 30293,
+ "ল": 30294,
+ "ŭ": 30295,
+ "午": 30296,
+ "ニ": 30297,
+ "\u0012": 30298,
+ "ʒ": 30299,
+ "შ": 30300,
+ "某": 30301,
+ "ォ": 30302,
+ "足": 30303,
+ "타": 30304,
+ "Ð": 30305,
+ "ხ": 30306,
+ "름": 30307,
+ "木": 30308,
+ "楼": 30309,
+ "최": 30310,
+ "红": 30311,
+ "¨": 30312,
+ "古": 30313,
+ "\u0006": 30314,
+ "단": 30315,
+ "今": 30316,
+ "ʔ": 30317,
+ "ट": 30318,
+ "ম": 30319,
+ "斯": 30320,
+ "語": 30321,
+ "Ÿ": 30322,
+ "🙄": 30323,
+ "牌": 30324,
+ "안": 30325,
+ "ស": 30326,
+ "颜": 30327,
+ "~": 30328,
+ "克": 30329,
+ "深": 30330,
+ "금": 30331,
+ "會": 30332,
+ "尔": 30333,
+ "释": 30334,
+ "批": 30335,
+ "산": 30336,
+ "野": 30337,
+ "防": 30338,
+ "Η": 30339,
+ "ө": 30340,
+ "ψ": 30341,
+ "ボ": 30342,
+ "": 30343,
+ "各": 30344,
+ "진": 30345,
+ "追": 30346,
+ "句": 30347,
+ "警": 30348,
+ "Φ": 30349,
+ "ѣ": 30350,
+ "ḍ": 30351,
+ "词": 30352,
+ "男": 30353,
+ "글": 30354,
+ "식": 30355,
+ "隐": 30356,
+ "복": 30357,
+ "盘": 30358,
+ "Ì": 30359,
+ "申": 30360,
+ "议": 30361,
+ "ザ": 30362,
+ "近": 30363,
+ "능": 30364,
+ "য": 30365,
+ "東": 30366,
+ "這": 30367,
+ "ர": 30368,
+ "距": 30369,
+ "院": 30370,
+ "德": 30371,
+ "ǐ": 30372,
+ "针": 30373,
+ "▀": 30374,
+ "↔": 30375,
+ "房": 30376,
+ "青": 30377,
+ "政": 30378,
+ "😅": 30379,
+ "递": 30380,
+ "প": 30381,
+ "波": 30382,
+ "ソ": 30383,
+ "绑": 30384,
+ "ビ": 30385,
+ "ễ": 30386,
+ "포": 30387,
+ "\u0010": 30388,
+ "ử": 30389,
+ "등": 30390,
+ "환": 30391,
+ "士": 30392,
+ "ত": 30393,
+ "Θ": 30394,
+ "초": 30395,
+ "境": 30396,
+ "差": 30397,
+ "采": 30398,
+ "디": 30399,
+ "ĩ": 30400,
+ "升": 30401,
+ "背": 30402,
+ "배": 30403,
+ "龙": 30404,
+ "街": 30405,
+ "್": 30406,
+ "ṛ": 30407,
+ "ু": 30408,
+ "弹": 30409,
+ "魔": 30410,
+ "객": 30411,
+ "‰": 30412,
+ "⌁": 30413,
+ "ἐ": 30414,
+ "禁": 30415,
+ "ผ": 30416,
+ "қ": 30417,
+ "島": 30418,
+ "ா": 30419,
+ "♭": 30420,
+ "百": 30421,
+ "ứ": 30422,
+ "ネ": 30423,
+ "专": 30424,
+ "來": 30425,
+ "刷": 30426,
+ "필": 30427,
+ "յ": 30428,
+ "ắ": 30429,
+ "华": 30430,
+ "Β": 30431,
+ "श": 30432,
+ "¸": 30433,
+ "屏": 30434,
+ "死": 30435,
+ "遍": 30436,
+ "검": 30437,
+ "Χ": 30438,
+ "것": 30439,
+ "八": 30440,
+ "览": 30441,
+ "택": 30442,
+ "唯": 30443,
+ "∙": 30444,
+ "¤": 30445,
+ "페": 30446,
+ "让": 30447,
+ "锁": 30448,
+ "무": 30449,
+ "思": 30450,
+ "隔": 30451,
+ "Ô": 30452,
+ "\u0013": 30453,
+ "ṃ": 30454,
+ "ワ": 30455,
+ "低": 30456,
+ "션": 30457,
+ "半": 30458,
+ "较": 30459,
+ "ត": 30460,
+ "享": 30461,
+ "积": 30462,
+ "": 30463,
+ "😊": 30464,
+ "典": 30465,
+ "ǔ": 30466,
+ "六": 30467,
+ "便": 30468,
+ "ɐ": 30469,
+ "简": 30470,
+ "继": 30471,
+ "仅": 30472,
+ "尾": 30473,
+ "": 30474,
+ "வ": 30475,
+ "կ": 30476,
+ "": 30477,
+ "영": 30478,
+ "火": 30479,
+ "湖": 30480,
+ "書": 30481,
+ "발": 30482,
+ "ハ": 30483,
+ "循": 30484,
+ "术": 30485,
+ "結": 30486,
+ "ļ": 30487,
+ "乐": 30488,
+ "滤": 30489,
+ "종": 30490,
+ "ถ": 30491,
+ "ὶ": 30492,
+ "满": 30493,
+ "╝": 30494,
+ "わ": 30495,
+ "ど": 30496,
+ "็": 30497,
+ "형": 30498,
+ "國": 30499,
+ "ự": 30500,
+ "線": 30501,
+ "블": 30502,
+ "封": 30503,
+ "確": 30504,
+ "依": 30505,
+ "ս": 30506,
+ "永": 30507,
+ "색": 30508,
+ "歌": 30509,
+ "數": 30510,
+ "福": 30511,
+ "삭": 30512,
+ "実": 30513,
+ "레": 30514,
+ "ſ": 30515,
+ "千": 30516,
+ "\u000e": 30517,
+ "母": 30518,
+ "더": 30519,
+ "임": 30520,
+ "տ": 30521,
+ "ے": 30522,
+ "几": 30523,
+ "双": 30524,
+ "노": 30525,
+ "ณ": 30526,
+ "掉": 30527,
+ "Ρ": 30528,
+ "ἀ": 30529,
+ "標": 30530,
+ "長": 30531,
+ "档": 30532,
+ "태": 30533,
+ "ペ": 30534,
+ "본": 30535,
+ "": 30536,
+ "底": 30537,
+ "终": 30538,
+ "請": 30539,
+ "კ": 30540,
+ "̯": 30541,
+ "예": 30542,
+ "▬": 30543,
+ "報": 30544,
+ "ピ": 30545,
+ "๏": 30546,
+ "暂": 30547,
+ "李": 30548,
+ "Υ": 30549,
+ "\u0005": 30550,
+ "\u0002": 30551,
+ "替": 30552,
+ "운": 30553,
+ "射": 30554,
+ "\u0018": 30555,
+ "매": 30556,
+ "\u0011": 30557,
+ "🏼": 30558,
+ "票": 30559,
+ "附": 30560,
+ "ノ": 30561,
+ "ũ": 30562,
+ "压": 30563,
+ "阿": 30564,
+ "Ò": 30565,
+ "테": 30566,
+ "∼": 30567,
+ "万": 30568,
+ "մ": 30569,
+ "후": 30570,
+ "普": 30571,
+ "截": 30572,
+ "속": 30573,
+ "括": 30574,
+ "😀": 30575,
+ "ை": 30576,
+ "▶": 30577,
+ "까": 30578,
+ "ট": 30579,
+ "曲": 30580,
+ "师": 30581,
+ "钱": 30582,
+ "栏": 30583,
+ "Ы": 30584,
+ "走": 30585,
+ "ữ": 30586,
+ "": 30587,
+ "归": 30588,
+ "점": 30589,
+ "🔥": 30590,
+ "었": 30591,
+ "連": 30592,
+ "私": 30593,
+ "청": 30594,
+ "刘": 30595,
+ "免": 30596,
+ "": 30597,
+ "奖": 30598,
+ "見": 30599,
+ "ֹ": 30600,
+ "☺": 30601,
+ "ケ": 30602,
+ "역": 30603,
+ "际": 30604,
+ "받": 30605,
+ "望": 30606,
+ "帝": 30607,
+ "减": 30608,
+ "두": 30609,
+ "领": 30610,
+ "": 30611,
+ "钟": 30612,
+ "ガ": 30613,
+ "架": 30614,
+ "든": 30615,
+ "ல": 30616,
+ "松": 30617,
+ "□": 30618,
+ "越": 30619,
+ "答": 30620,
+ "ɕ": 30621,
+ "ῦ": 30622,
+ "染": 30623,
+ "": 30624,
+ "质": 30625,
+ "顺": 30626,
+ "气": 30627,
+ "╗": 30628,
+ "計": 30629,
+ "ქ": 30630,
+ "亮": 30631,
+ "🤦": 30632,
+ "̂": 30633,
+ "ٹ": 30634,
+ "座": 30635,
+ "ˌ": 30636,
+ "均": 30637,
+ "\u000b": 30638,
+ "官": 30639,
+ "适": 30640,
+ "护": 30641,
+ "久": 30642,
+ "春": 30643,
+ "曹": 30644,
+ "皇": 30645,
+ "脚": 30646,
+ "池": 30647,
+ "延": 30648,
+ "키": 30649,
+ "품": 30650,
+ "現": 30651,
+ "檔": 30652,
+ "ば": 30653,
+ "ⴰ": 30654,
+ "希": 30655,
+ "玩": 30656,
+ "固": 30657,
+ "黄": 30658,
+ "": 30659,
+ "☽": 30660,
+ "银": 30661,
+ "\u0003": 30662,
+ "┃": 30663,
+ "👏": 30664,
+ "불": 30665,
+ "攻": 30666,
+ "へ": 30667,
+ "决": 30668,
+ "⊙": 30669,
+ "宁": 30670,
+ "च": 30671,
+ "機": 30672,
+ "義": 30673,
+ "ɲ": 30674,
+ "\u0015": 30675,
+ "했": 30676,
+ "ẩ": 30677,
+ "愛": 30678,
+ "矩": 30679,
+ "패": 30680,
+ "ặ": 30681,
+ "郎": 30682,
+ "Ь": 30683,
+ "绘": 30684,
+ "负": 30685,
+ "ổ": 30686,
+ "ய": 30687,
+ "汉": 30688,
+ "編": 30689,
+ "ێ": 30690,
+ "്": 30691,
+ "じ": 30692,
+ "카": 30693,
+ "似": 30694,
+ "ں": 30695,
+ "や": 30696,
+ "認": 30697,
+ "\u000f": 30698,
+ "過": 30699,
+ "통": 30700,
+ "▪": 30701,
+ "约": 30702,
+ "香": 30703,
+ "买": 30704,
+ "住": 30705,
+ "╚": 30706,
+ "😁": 30707,
+ "扩": 30708,
+ "静": 30709,
+ "려": 30710,
+ "학": 30711,
+ "钥": 30712,
+ "증": 30713,
+ "ỉ": 30714,
+ "她": 30715,
+ "食": 30716,
+ "往": 30717,
+ "點": 30718,
+ "偏": 30719,
+ "康": 30720,
+ "\u0014": 30721,
+ "į": 30722,
+ "준": 30723,
+ "\u0004": 30724,
+ "ฟ": 30725,
+ "♣": 30726,
+ "戏": 30727,
+ "ʂ": 30728,
+ "井": 30729,
+ "军": 30730,
+ "爱": 30731,
+ "ٱ": 30732,
+ "七": 30733,
+ "차": 30734,
+ "币": 30735,
+ "♠": 30736,
+ "哈": 30737,
+ "阅": 30738,
+ "介": 30739,
+ "观": 30740,
+ "區": 30741,
+ "˜": 30742,
+ "ً": 30743,
+ "又": 30744,
+ "冲": 30745,
+ "朝": 30746,
+ "姓": 30747,
+ "课": 30748,
+ "龍": 30749,
+ "각": 30750,
+ "∈": 30751,
+ "米": 30752,
+ "ƒ": 30753,
+ "喜": 30754,
+ "夜": 30755,
+ "团": 30756,
+ "⇒": 30757,
+ "远": 30758,
+ "\u001a": 30759,
+ "ὐ": 30760,
+ "承": 30761,
+ "ಿ": 30762,
+ "室": 30763,
+ "ʀ": 30764,
+ "ង": 30765,
+ "अ": 30766,
+ "罗": 30767,
+ "🙏": 30768,
+ "软": 30769,
+ "🟡": 30770,
+ "건": 30771,
+ "؟": 30772,
+ "း": 30773,
+ "ᴇ": 30774,
+ "ユ": 30775,
+ "토": 30776,
+ "策": 30777,
+ "̄": 30778,
+ "국": 30779,
+ "ֶ": 30780,
+ "协": 30781,
+ "营": 30782,
+ "関": 30783,
+ "吉": 30784,
+ "💀": 30785,
+ "奇": 30786,
+ "滚": 30787,
+ "轴": 30788,
+ "処": 30789,
+ "土": 30790,
+ "划": 30791,
+ "ड": 30792,
+ "临": 30793,
+ "ֵ": 30794,
+ "航": 30795,
+ "浏": 30796,
+ "ゴ": 30797,
+ "別": 30798,
+ "寺": 30799,
+ "於": 30800,
+ "進": 30801,
+ "ὸ": 30802,
+ "風": 30803,
+ "ன": 30804,
+ "班": 30805,
+ "◼": 30806,
+ "九": 30807,
+ "̥": 30808,
+ "號": 30809,
+ "류": 30810,
+ "础": 30811,
+ "般": 30812,
+ "︙": 30813,
+ "̈": 30814,
+ "番": 30815,
+ "✨": 30816,
+ "😎": 30817,
+ "ো": 30818,
+ "😍": 30819,
+ "單": 30820,
+ "帧": 30821,
+ "授": 30822,
+ "赋": 30823,
+ "巴": 30824,
+ "占": 30825,
+ "假": 30826,
+ "ṅ": 30827,
+ "透": 30828,
+ "項": 30829,
+ "ħ": 30830,
+ "馬": 30831,
+ "🟢": 30832,
+ "Ľ": 30833,
+ "լ": 30834,
+ "券": 30835,
+ "같": 30836,
+ "類": 30837,
+ "對": 30838,
+ "월": 30839,
+ "激": 30840,
+ "\u0017": 30841,
+ "戦": 30842,
+ "独": 30843,
+ "訊": 30844,
+ "ិ": 30845,
+ "套": 30846,
+ "ʷ": 30847,
+ "跟": 30848,
+ "ở": 30849,
+ "渲": 30850,
+ "顯": 30851,
+ "降": 30852,
+ "ာ": 30853,
+ "尼": 30854,
+ "血": 30855,
+ "언": 30856,
+ "牛": 30857,
+ "將": 30858,
+ "ศ": 30859,
+ "拍": 30860,
+ "刻": 30861,
+ "ზ": 30862,
+ "╔": 30863,
+ "藤": 30864,
+ "్": 30865,
+ "ῶ": 30866,
+ "🟠": 30867,
+ "良": 30868,
+ "김": 30869,
+ "দ": 30870,
+ "Ṣ": 30871,
+ "録": 30872,
+ "伊": 30873,
+ "落": 30874,
+ "雄": 30875,
+ "雪": 30876,
+ "映": 30877,
+ "著": 30878,
+ "른": 30879,
+ "ფ": 30880,
+ "対": 30881,
+ "智": 30882,
+ "译": 30883,
+ "┬": 30884,
+ "抽": 30885,
+ "ῖ": 30886,
+ "酒": 30887,
+ "Ћ": 30888,
+ "股": 30889,
+ "់": 30890,
+ "순": 30891,
+ "직": 30892,
+ "भ": 30893,
+ "谷": 30894,
+ "물": 30895,
+ "ǒ": 30896,
+ "⠄": 30897,
+ "热": 30898,
+ "終": 30899,
+ "夹": 30900,
+ "干": 30901,
+ "彩": 30902,
+ "敗": 30903,
+ "ќ": 30904,
+ "♯": 30905,
+ "̣": 30906,
+ "վ": 30907,
+ "轮": 30908,
+ "阵": 30909,
+ "夏": 30910,
+ "幕": 30911,
+ "吧": 30912,
+ "港": 30913,
+ "益": 30914,
+ "儿": 30915,
+ "액": 30916,
+ "售": 30917,
+ "兵": 30918,
+ "惠": 30919,
+ "欢": 30920,
+ "": 30921,
+ "零": 30922,
+ "學": 30923,
+ "": 30924,
+ "員": 30925,
+ "ỗ": 30926,
+ "玉": 30927,
+ "逻": 30928,
+ "᥀": 30929,
+ "吗": 30930,
+ "沒": 30931,
+ "≠": 30932,
+ "너": 30933,
+ "ச": 30934,
+ "\u0016": 30935,
+ "夫": 30936,
+ "წ": 30937,
+ "堂": 30938,
+ "電": 30939,
+ "≡": 30940,
+ "陆": 30941,
+ "져": 30942,
+ "研": 30943,
+ "荐": 30944,
+ "健": 30945,
+ "碼": 30946,
+ "练": 30947,
+ "検": 30948,
+ "송": 30949,
+ "ै": 30950,
+ "哪": 30951,
+ "圆": 30952,
+ "Ա": 30953,
+ "↩": 30954,
+ "托": 30955,
+ "̪": 30956,
+ "ू": 30957,
+ "缀": 30958,
+ "네": 30959,
+ "沙": 30960,
+ "兴": 30961,
+ "病": 30962,
+ "\u0007": 30963,
+ "ល": 30964,
+ "ừ": 30965,
+ "Ἀ": 30966,
+ "강": 30967,
+ "항": 30968,
+ "\u0019": 30969,
+ "換": 30970,
+ "温": 30971,
+ "帖": 30972,
+ "ទ": 30973,
+ "込": 30974,
+ "削": 30975,
+ "알": 30976,
+ "征": 30977,
+ "习": 30978,
+ "법": 30979,
+ "栈": 30980,
+ "绝": 30981,
+ "": 30982,
+ "ڕ": 30983,
+ "圖": 30984,
+ "苏": 30985,
+ "発": 30986,
+ "ု": 30987,
+ "町": 30988,
+ "互": 30989,
+ "়": 30990,
+ "ც": 30991,
+ "守": 30992,
+ "새": 30993,
+ "侧": 30994,
+ "草": 30995,
+ "ས": 30996,
+ "扫": 30997,
+ "‒": 30998,
+ "恢": 30999,
+ "ң": 31000,
+ "ण": 31001,
+ "ற": 31002,
+ "째": 31003,
+ "්": 31004,
+ "拟": 31005,
+ "派": 31006,
+ "🏽": 31007,
+ "呼": 31008,
+ "": 31009,
+ "演": 31010,
+ "究": 31011,
+ "교": 31012,
+ "ɣ": 31013,
+ "ए": 31014,
+ "ី": 31015,
+ "ף": 31016,
+ "富": 31017,
+ "駅": 31018,
+ "ず": 31019,
+ "♪": 31020,
+ "😆": 31021,
+ "접": 31022,
+ "ғ": 31023,
+ "▓": 31024,
+ "존": 31025,
+ "ಾ": 31026,
+ "旋": 31027,
+ "ゃ": 31028,
+ "补": 31029,
+ "ץ": 31030,
+ "門": 31031,
+ "ច": 31032,
+ "날": 31033,
+ "ภ": 31034,
+ "ག": 31035,
+ "傳": 31036,
+ "∆": 31037,
+ "": 31038,
+ "ׁ": 31039,
+ "缺": 31040,
+ "頭": 31041,
+ "怪": 31042,
+ "組": 31043,
+ "별": 31044,
+ "Ъ": 31045,
+ "發": 31046,
+ "雷": 31047,
+ "ರ": 31048,
+ "ซ": 31049,
+ "び": 31050,
+ "翻": 31051,
+ "ھ": 31052,
+ "პ": 31053,
+ "題": 31054,
+ "居": 31055,
+ "집": 31056,
+ "🌍": 31057,
+ "˚": 31058,
+ "避": 31059,
+ "줄": 31060,
+ "ុ": 31061,
+ "滑": 31062,
+ "故": 31063,
+ "ญ": 31064,
+ "〜": 31065,
+ "ನ": 31066,
+ "양": 31067,
+ "완": 31068,
+ "ள": 31069,
+ "倍": 31070,
+ "宗": 31071,
+ "択": 31072,
+ "브": 31073,
+ "ɴ": 31074,
+ "効": 31075,
+ "尺": 31076,
+ "視": 31077,
+ "ẽ": 31078,
+ "覆": 31079,
+ "ध": 31080,
+ "骨": 31081,
+ "달": 31082,
+ "ᴛ": 31083,
+ "蓝": 31084,
+ "關": 31085,
+ "額": 31086,
+ "Õ": 31087,
+ "∗": 31088,
+ "卷": 31089,
+ "갑": 31090,
+ "르": 31091,
+ "众": 31092,
+ "ᴀ": 31093,
+ "態": 31094,
+ "ٰ": 31095,
+ "暗": 31096,
+ "君": 31097,
+ "錯": 31098,
+ "ɒ": 31099,
+ "យ": 31100,
+ "ḫ": 31101,
+ "ῆ": 31102,
+ "亚": 31103,
+ "♡": 31104,
+ "割": 31105,
+ "鼠": 31106,
+ "̶": 31107,
+ "Ë": 31108,
+ "読": 31109,
+ "격": 31110,
+ "ゲ": 31111,
+ "眼": 31112,
+ "Ý": 31113,
+ "ژ": 31114,
+ "雨": 31115,
+ "宮": 31116,
+ "쪽": 31117,
+ "ष": 31118,
+ "複": 31119,
+ "剩": 31120,
+ "早": 31121,
+ "杂": 31122,
+ "焦": 31123,
+ "贝": 31124,
+ "突": 31125,
+ "워": 31126,
+ "另": 31127,
+ "摄": 31128,
+ "\b": 31129,
+ "": 31130,
+ "府": 31131,
+ "외": 31132,
+ "盖": 31133,
+ "\u001c": 31134,
+ "ษ": 31135,
+ "佛": 31136,
+ "概": 31137,
+ "與": 31138,
+ "經": 31139,
+ "-": 31140,
+ "һ": 31141,
+ "問": 31142,
+ "ು": 31143,
+ "ἰ": 31144,
+ "話": 31145,
+ "倒": 31146,
+ "葛": 31147,
+ "べ": 31148,
+ "ろ": 31149,
+ "\u001e": 31150,
+ "।": 31151,
+ "ေ": 31152,
+ "ᴏ": 31153,
+ "训": 31154,
+ "體": 31155,
+ "👌": 31156,
+ "內": 31157,
+ "က": 31158,
+ "企": 31159,
+ "약": 31160,
+ "찾": 31161,
+ "ོ": 31162,
+ "破": 31163,
+ "輸": 31164,
+ "림": 31165,
+ "塔": 31166,
+ "턴": 31167,
+ "杀": 31168,
+ "』": 31169,
+ "味": 31170,
+ "浮": 31171,
+ "┆": 31172,
+ "ġ": 31173,
+ "郡": 31174,
+ "┐": 31175,
+ "『": 31176,
+ "阶": 31177,
+ "雅": 31178,
+ "┈": 31179,
+ "园": 31180,
+ ".": 31181,
+ "吃": 31182,
+ "남": 31183,
+ " ": 31184,
+ "ར": 31185,
+ "帮": 31186,
+ "毛": 31187,
+ "耗": 31188,
+ "举": 31189,
+ "ర": 31190,
+ "拿": 31191,
+ "밀": 31192,
+ "ご": 31193,
+ "够": 31194,
+ "礼": 31195,
+ "ព": 31196,
+ "ね": 31197,
+ "": 31198,
+ "兰": 31199,
+ "❌": 31200,
+ "折": 31201,
+ "십": 31202,
+ "💎": 31203,
+ "業": 31204,
+ "诸": 31205,
+ "孙": 31206,
+ "བ": 31207,
+ "😳": 31208,
+ "種": 31209,
+ "Ï": 31210,
+ "ึ": 31211,
+ "": 31212,
+ "医": 31213,
+ "拼": 31214,
+ "↵": 31215,
+ "⅓": 31216,
+ "\u001f": 31217,
+ "မ": 31218,
+ "叫": 31219,
+ "জ": 31220,
+ "予": 31221,
+ "寸": 31222,
+ "梅": 31223,
+ "醒": 31224,
+ "津": 31225,
+ "န": 31226,
+ "ి": 31227,
+ "厂": 31228,
+ "屋": 31229,
+ "ख": 31230,
+ "師": 31231,
+ "👀": 31232,
+ "ỏ": 31233,
+ "ヤ": 31234,
+ "ὰ": 31235,
+ "\u001d": 31236,
+ "◆": 31237,
+ "ដ": 31238,
+ "材": 31239,
+ "ホ": 31240,
+ "張": 31241,
+ "洞": 31242,
+ "餐": 31243,
+ "천": 31244,
+ "হ": 31245,
+ "達": 31246,
+ "們": 31247,
+ "斗": 31248,
+ "横": 31249,
+ "백": 31250,
+ "ំ": 31251,
+ "ۆ": 31252,
+ "말": 31253,
+ "গ": 31254,
+ "佳": 31255,
+ "랜": 31256,
+ "仁": 31257,
+ "陈": 31258,
+ "飞": 31259,
+ "极": 31260,
+ "": 31261,
+ "및": 31262,
+ "仓": 31263,
+ "⬛": 31264,
+ "昌": 31265,
+ "錢": 31266,
+ "殊": 31267,
+ "┴": 31268,
+ "○": 31269,
+ "길": 31270,
+ "泉": 31271,
+ "甲": 31272,
+ "활": 31273,
+ "ひ": 31274,
+ "শ": 31275,
+ "ን": 31276,
+ "Ť": 31277,
+ "ღ": 31278,
+ "皮": 31279,
+ "強": 31280,
+ "赛": 31281,
+ "ా": 31282,
+ "預": 31283,
+ "င": 31284,
+ "튼": 31285,
+ "플": 31286,
+ "ყ": 31287,
+ "⋆": 31288,
+ "ք": 31289,
+ "ા": 31290,
+ "尚": 31291,
+ "또": 31292,
+ "բ": 31293,
+ "┌": 31294,
+ "節": 31295,
+ "森": 31296,
+ "आ": 31297,
+ "办": 31298,
+ "園": 31299,
+ "牙": 31300,
+ "庆": 31301,
+ "隆": 31302,
+ "😔": 31303,
+ "叉": 31304,
+ "գ": 31305,
+ "피": 31306,
+ "ギ": 31307,
+ "啊": 31308,
+ "続": 31309,
+ "灵": 31310,
+ "ヒ": 31311,
+ "忽": 31312,
+ "ʌ": 31313,
+ "량": 31314,
+ "油": 31315,
+ "讯": 31316,
+ "ⵉ": 31317,
+ "릭": 31318,
+ "刚": 31319,
+ "氏": 31320,
+ "ိ": 31321,
+ "Ī": 31322,
+ "誤": 31323,
+ "齐": 31324,
+ "末": 31325,
+ "🙌": 31326,
+ "̞": 31327,
+ "圈": 31328,
+ "念": 31329,
+ "숫": 31330,
+ "毫": 31331,
+ "當": 31332,
+ "規": 31333,
+ "판": 31334,
+ "ు": 31335,
+ "旧": 31336,
+ "卖": 31337,
+ "ฉ": 31338,
+ "幸": 31339,
+ "署": 31340,
+ "근": 31341,
+ "ই": 31342,
+ "岛": 31343,
+ "դ": 31344,
+ "觉": 31345,
+ "害": 31346,
+ "毕": 31347,
+ "ฐ": 31348,
+ "威": 31349,
+ "育": 31350,
+ "呢": 31351,
+ "峰": 31352,
+ "职": 31353,
+ "陽": 31354,
+ "ි": 31355,
+ "亞": 31356,
+ "ұ": 31357,
+ "₃": 31358,
+ "따": 31359,
+ "施": 31360,
+ "泰": 31361,
+ "載": 31362,
+ "
": 31363,
+ "笑": 31364,
+ "華": 31365,
+ "迎": 31366,
+ "됩": 31367,
+ "豆": 31368,
+ "嘉": 31369,
+ "🤡": 31370,
+ "ĕ": 31371,
+ "庄": 31372,
+ "級": 31373,
+ "Ψ": 31374,
+ "ི": 31375,
+ "気": 31376,
+ "责": 31377,
+ "հ": 31378,
+ "អ": 31379,
+ "乱": 31380,
+ "休": 31381,
+ "約": 31382,
+ "ฆ": 31383,
+ "∑": 31384,
+ "察": 31385,
+ "온": 31386,
+ "😬": 31387,
+ "ড": 31388,
+ "乘": 31389,
+ "람": 31390,
+ "इ": 31391,
+ "Ά": 31392,
+ "ந": 31393,
+ "ើ": 31394,
+ "亲": 31395,
+ "េ": 31396,
+ "委": 31397,
+ "赤": 31398,
+ "됨": 31399,
+ "勝": 31400,
+ "怎": 31401,
+ "감": 31402,
+ "宋": 31403,
+ "調": 31404,
+ "짜": 31405,
+ "ী": 31406,
+ "难": 31407,
+ "못": 31408,
+ "티": 31409,
+ "備": 31410,
+ "塞": 31411,
+ "វ": 31412,
+ "险": 31413,
+ "旅": 31414,
+ "虚": 31415,
+ "↳": 31416,
+ "笔": 31417,
+ "馆": 31418,
+ "Қ": 31419,
+ "⚡": 31420,
+ "ೆ": 31421,
+ "※": 31422,
+ "唐": 31423,
+ "律": 31424,
+ "稍": 31425,
+ "散": 31426,
+ "ર": 31427,
+ "ヴ": 31428,
+ "副": 31429,
+ "尽": 31430,
+ "挂": 31431,
+ "県": 31432,
+ "⚠": 31433,
+ "洋": 31434,
+ "鬼": 31435,
+ "암": 31436,
+ "孩": 31437,
+ "℃": 31438,
+ "並": 31439,
+ "ց": 31440,
+ "ូ": 31441,
+ "ℓ": 31442,
+ "ⵏ": 31443,
+ "扣": 31444,
+ "铁": 31445,
+ "闻": 31446,
+ "ˆ": 31447,
+ "戳": 31448,
+ "む": 31449,
+ "秀": 31450,
+ "細": 31451,
+ "ပ": 31452,
+ "御": 31453,
+ "拖": 31454,
+ "좌": 31455,
+ "ؤ": 31456,
+ "绍": 31457,
+ "ỹ": 31458,
+ "참": 31459,
+ "향": 31460,
+ "Ď": 31461,
+ "끝": 31462,
+ "민": 31463,
+ "ძ": 31464,
+ "贵": 31465,
+ "纪": 31466,
+ "秋": 31467,
+ "ಕ": 31468,
+ "ӏ": 31469,
+ "網": 31470,
+ "铺": 31471,
+ "恋": 31472,
+ "fl": 31473,
+ "兼": 31474,
+ "羽": 31475,
+ "창": 31476,
+ "啟": 31477,
+ "弟": 31478,
+ "년": 31479,
+ "慢": 31480,
+ "효": 31481,
+ "許": 31482,
+ "硬": 31483,
+ "잘": 31484,
+ "템": 31485,
+ "્": 31486,
+ "න": 31487,
+ "術": 31488,
+ "ڈ": 31489,
+ "溪": 31490,
+ "": 31491,
+ "暴": 31492,
+ "混": 31493,
+ "夢": 31494,
+ "랑": 31495,
+ "আ": 31496,
+ "還": 31497,
+ "探": 31498,
+ "祖": 31499,
+ "织": 31500,
+ "軍": 31501,
+ "թ": 31502,
+ "務": 31503,
+ "艺": 31504,
+ "ད": 31505,
+ "ት": 31506,
+ "ṁ": 31507,
+ "應": 31508,
+ "擇": 31509,
+ "🥰": 31510,
+ "ķ": 31511,
+ "渡": 31512,
+ "葉": 31513,
+ "령": 31514,
+ "決": 31515,
+ "刀": 31516,
+ "從": 31517,
+ "變": 31518,
+ "올": 31519,
+ "💪": 31520,
+ "灣": 31521,
+ "ር": 31522,
+ "평": 31523,
+ "衣": 31524,
+ "😄": 31525,
+ "ി": 31526,
+ "ჩ": 31527,
+ "ὁ": 31528,
+ "ほ": 31529,
+ "Û": 31530,
+ "চ": 31531,
+ "ර": 31532,
+ "製": 31533,
+ "隊": 31534,
+ "₱": 31535,
+ "纳": 31536,
+ "赖": 31537,
+ "农": 31538,
+ "桥": 31539,
+ "ỳ": 31540,
+ "🏾": 31541,
+ "阻": 31542,
+ "ជ": 31543,
+ "秘": 31544,
+ "박": 31545,
+ "伤": 31546,
+ "稿": 31547,
+ "ం": 31548,
+ "拦": 31549,
+ "넣": 31550,
+ "💕": 31551,
+ "₁": 31552,
+ "宿": 31553,
+ "錄": 31554,
+ "镜": 31555,
+ "채": 31556,
+ "Ə": 31557,
+ "ང": 31558,
+ "⇔": 31559,
+ "☼": 31560,
+ "ུ": 31561,
+ "党": 31562,
+ "급": 31563,
+ "洲": 31564,
+ "ղ": 31565,
+ "說": 31566,
+ "ĭ": 31567,
+ "尝": 31568,
+ "담": 31569,
+ "फ": 31570,
+ "哥": 31571,
+ "圣": 31572,
+ "萨": 31573,
+ "😏": 31574,
+ "ʏ": 31575,
+ "ெ": 31576,
+ "丁": 31577,
+ "虎": 31578,
+ "권": 31579,
+ "善": 31580,
+ "岩": 31581,
+ "커": 31582,
+ "◦": 31583,
+ "抛": 31584,
+ "석": 31585,
+ "Έ": 31586,
+ "宣": 31587,
+ "拳": 31588,
+ "팅": 31589,
+ "枚": 31590,
+ "洛": 31591,
+ "証": 31592,
+ "陵": 31593,
+ "佐": 31594,
+ "館": 31595,
+ "누": 31596,
+ "돌": 31597,
+ "₄": 31598,
+ "稱": 31599,
+ "聊": 31600,
+ "車": 31601,
+ "루": 31602,
+ "״": 31603,
+ "ಠ": 31604,
+ "庫": 31605,
+ "མ": 31606,
+ "統": 31607,
+ "련": 31608,
+ "़": 31609,
+ "ṯ": 31610,
+ "ക": 31611,
+ "旗": 31612,
+ "励": 31613,
+ "紀": 31614,
+ "忠": 31615,
+ "າ": 31616,
+ "杨": 31617,
+ "丹": 31618,
+ "Ù": 31619,
+ "ฝ": 31620,
+ "却": 31621,
+ "舞": 31622,
+ "轉": 31623,
+ "တ": 31624,
+ "丽": 31625,
+ "借": 31626,
+ "ා": 31627,
+ "ょ": 31628,
+ "옵": 31629,
+ "편": 31630,
+ "蒙": 31631,
+ "衡": 31632,
+ "ʋ": 31633,
+ "叶": 31634,
+ "̇": 31635,
+ "⬜": 31636,
+ "🇺": 31637,
+ "Հ": 31638,
+ "谢": 31639,
+ "Ą": 31640,
+ "ே": 31641,
+ "ằ": 31642,
+ "既": 31643,
+ "济": 31644,
+ "≯": 31645,
+ "準": 31646,
+ "답": 31647,
+ "ಲ": 31648,
+ "残": 31649,
+ "虑": 31650,
+ "̆": 31651,
+ "┘": 31652,
+ "急": 31653,
+ "招": 31654,
+ "막": 31655,
+ "≮": 31656,
+ "產": 31657,
+ "Ṭ": 31658,
+ "😢": 31659,
+ "垂": 31660,
+ "親": 31661,
+ "ģ": 31662,
+ "־": 31663,
+ "猫": 31664,
+ "ʟ": 31665,
+ "☃": 31666,
+ "✪": 31667,
+ "刪": 31668,
+ "胡": 31669,
+ "☉": 31670,
+ "晚": 31671,
+ "군": 31672,
+ "승": 31673,
+ "న": 31674,
+ "ὴ": 31675,
+ "曾": 31676,
+ "論": 31677,
+ "ɯ": 31678,
+ "త": 31679,
+ "戰": 31680,
+ "鱼": 31681,
+ "ǧ": 31682,
+ "寶": 31683,
+ "특": 31684,
+ "💯": 31685,
+ "崎": 31686,
+ "甘": 31687,
+ "該": 31688,
+ "링": 31689,
+ "😡": 31690,
+ "उ": 31691,
+ "ែ": 31692,
+ "頁": 31693,
+ "큰": 31694,
+ "➤": 31695,
+ "총": 31696,
+ "💰": 31697,
+ "∂": 31698,
+ "毁": 31699,
+ "聖": 31700,
+ "麻": 31701,
+ "ʐ": 31702,
+ "敏": 31703,
+ "運": 31704,
+ "될": 31705,
+ "쓰": 31706,
+ "ಸ": 31707,
+ "စ": 31708,
+ "✦": 31709,
+ "젝": 31710,
+ "復": 31711,
+ "寻": 31712,
+ "茶": 31713,
+ "ਾ": 31714,
+ "竹": 31715,
+ "遇": 31716,
+ "順": 31717,
+ "며": 31718,
+ "累": 31719,
+ "ĝ": 31720,
+ "ˇ": 31721,
+ "覧": 31722,
+ "এ": 31723,
+ "株": 31724,
+ "취": 31725,
+ "ስ": 31726,
+ "争": 31727,
+ "势": 31728,
+ "宇": 31729,
+ "橋": 31730,
+ "Ӏ": 31731,
+ "堆": 31732,
+ "ⵙ": 31733,
+ "丶": 31734,
+ "棋": 31735,
+ "肉": 31736,
+ "የ": 31737,
+ "": 31738,
+ "❶": 31739,
+ "季": 31740,
+ "ል": 31741,
+ "殿": 31742,
+ "優": 31743,
+ "試": 31744,
+ "첫": 31745,
+ "Ό": 31746,
+ "戶": 31747,
+ "ண": 31748,
+ "羅": 31749,
+ "桃": 31750,
+ "립": 31751,
+ "浪": 31752,
+ "脑": 31753,
+ "😛": 31754,
+ "弃": 31755,
+ "炮": 31756,
+ "轻": 31757,
+ "울": 31758,
+ "": 31759,
+ "ヘ": 31760,
+ "奥": 31761,
+ "💜": 31762,
+ "忘": 31763,
+ "遠": 31764,
+ "飛": 31765,
+ "魏": 31766,
+ "Ē": 31767,
+ "汇": 31768,
+ "央": 31769,
+ "逆": 31770,
+ "露": 31771,
+ "須": 31772,
+ "ѐ": 31773,
+ "ḷ": 31774,
+ "ದ": 31775,
+ "✭": 31776,
+ "寄": 31777,
+ "盟": 31778,
+ "财": 31779,
+ "際": 31780,
+ "ἔ": 31781,
+ "ǫ": 31782,
+ "थ": 31783,
+ "ാ": 31784,
+ "宫": 31785,
+ "巨": 31786,
+ "途": 31787,
+ "ʹ": 31788,
+ "ಗ": 31789,
+ "帐": 31790,
+ "": 31791,
+ "拒": 31792,
+ "药": 31793,
+ "🙃": 31794,
+ "ŕ": 31795,
+ "亡": 31796,
+ "壁": 31797,
+ "ም": 31798,
+ "參": 31799,
+ "😩": 31800,
+ "շ": 31801,
+ "ವ": 31802,
+ "ណ": 31803,
+ "丰": 31804,
+ "獲": 31805,
+ "莉": 31806,
+ "좋": 31807,
+ "ရ": 31808,
+ "₦": 31809,
+ "겠": 31810,
+ "👉": 31811,
+ "吴": 31812,
+ "岡": 31813,
+ "诉": 31814,
+ "읽": 31815,
+ "🥺": 31816,
+ "爆": 31817,
+ "🇸": 31818,
+ "ভ": 31819,
+ "迭": 31820,
+ "엔": 31821,
+ "ἄ": 31822,
+ "捷": 31823,
+ "納": 31824,
+ "邀": 31825,
+ "ಯ": 31826,
+ "爾": 31827,
+ "船": 31828,
+ "赞": 31829,
+ "胜": 31830,
+ "므": 31831,
+ "သ": 31832,
+ "構": 31833,
+ "磁": 31834,
+ "冰": 31835,
+ "딩": 31836,
+ "ે": 31837,
+ "媒": 31838,
+ "繁": 31839,
+ "☠": 31840,
+ "❒": 31841,
+ "仪": 31842,
+ "렬": 31843,
+ "昭": 31844,
+ "珠": 31845,
+ "離": 31846,
+ "ན": 31847,
+ "ల": 31848,
+ "ತ": 31849,
+ "拷": 31850,
+ "粉": 31851,
+ "벤": 31852,
+ "⇽": 31853,
+ "乌": 31854,
+ "拥": 31855,
+ "ҳ": 31856,
+ "ය": 31857,
+ "ེ": 31858,
+ "仙": 31859,
+ "塊": 31860,
+ "幅": 31861,
+ "🎉": 31862,
+ "Մ": 31863,
+ "跨": 31864,
+ "ٔ": 31865,
+ "恩": 31866,
+ "损": 31867,
+ "养": 31868,
+ "奈": 31869,
+ "ǀ": 31870,
+ "严": 31871,
+ "卫": 31872,
+ "迟": 31873,
+ "様": 31874,
+ "裡": 31875,
+ "난": 31876,
+ "았": 31877,
+ "͜": 31878,
+ "Ζ": 31879,
+ "ਰ": 31880,
+ "պ": 31881,
+ "ং": 31882,
+ "丢": 31883,
+ "伝": 31884,
+ "컨": 31885,
+ "ව": 31886,
+ "ြ": 31887,
+ "冷": 31888,
+ "遗": 31889,
+ "銀": 31890,
+ "̌": 31891,
+ "ᴜ": 31892,
+ "瑞": 31893,
+ "ฌ": 31894,
+ "❍": 31895,
+ "ふ": 31896,
+ "聚": 31897,
+ "碎": 31898,
+ "衛": 31899,
+ "অ": 31900,
+ "ញ": 31901,
+ "퍼": 31902,
+ "Ս": 31903,
+ "ນ": 31904,
+ "ẓ": 31905,
+ "✌": 31906,
+ "孝": 31907,
+ "陳": 31908,
+ "히": 31909,
+ "ක": 31910,
+ "黒": 31911,
+ "💖": 31912,
+ "ḩ": 31913,
+ "応": 31914,
+ "饰": 31915,
+ "∪": 31916,
+ "宜": 31917,
+ "樂": 31918,
+ "則": 31919,
+ "勇": 31920,
+ "徐": 31921,
+ "ⵓ": 31922,
+ "權": 31923,
+ "鲁": 31924,
+ "‟": 31925,
+ "庭": 31926,
+ "苗": 31927,
+ "🔴": 31928,
+ "闲": 31929,
+ "독": 31930,
+ "ɹ": 31931,
+ "ҽ": 31932,
+ "ថ": 31933,
+ "宏": 31934,
+ "尊": 31935,
+ "總": 31936,
+ "裝": 31937,
+ "ම": 31938,
+ "▸": 31939,
+ "測": 31940,
+ "ಮ": 31941,
+ "አ": 31942,
+ "轩": 31943,
+ "兄": 31944,
+ "剑": 31945,
+ "ન": 31946,
+ "朱": 31947,
+ "ǝ": 31948,
+ "Ḩ": 31949,
+ "担": 31950,
+ "灰": 31951,
+ "讲": 31952,
+ "롤": 31953,
+ "︎": 31954,
+ "😤": 31955,
+ "ោ": 31956,
+ "애": 31957,
+ "였": 31958,
+ "질": 31959,
+ "振": 31960,
+ "灯": 31961,
+ "ĉ": 31962,
+ "ස": 31963,
+ "閉": 31964,
+ "램": 31965,
+ "ಂ": 31966,
+ "げ": 31967,
+ "̧": 31968,
+ "狂": 31969,
+ "融": 31970,
+ "仍": 31971,
+ "實": 31972,
+ "楽": 31973,
+ "範": 31974,
+ "ٌ": 31975,
+ "వ": 31976,
+ "嵌": 31977,
+ "摩": 31978,
+ "袁": 31979,
+ "ষ": 31980,
+ "乎": 31981,
+ "규": 31982,
+ "岗": 31983,
+ "糊": 31984,
+ "క": 31985,
+ "雲": 31986,
+ "심": 31987,
+ "ई": 31988,
+ "འ": 31989,
+ "ἡ": 31990,
+ "丝": 31991,
+ "Ħ": 31992,
+ "ٍ": 31993,
+ "ٓ": 31994,
+ "အ": 31995,
+ "執": 31996,
+ "벨": 31997,
+ "ゼ": 31998,
+ "梦": 31999
+ },
+ "merges": [
+ "▁ t",
+ "i n",
+ "e r",
+ "▁ a",
+ "h e",
+ "o n",
+ "r e",
+ "▁ s",
+ "e n",
+ "a t",
+ "o r",
+ "▁t he",
+ "▁th e",
+ "▁ the",
+ "e s",
+ "▁ w",
+ "a n",
+ "▁ c",
+ "i s",
+ "i t",
+ "o u",
+ "▁ d",
+ "a l",
+ "a r",
+ "▁ p",
+ "▁ f",
+ "e d",
+ "▁ b",
+ "in g",
+ "i ng",
+ "▁ o",
+ "▁ m",
+ "l e",
+ "n d",
+ "a s",
+ "i c",
+ "▁ h",
+ "io n",
+ "i on",
+ "▁i n",
+ "▁ in",
+ "▁t o",
+ "▁ to",
+ "e t",
+ "o m",
+ "e l",
+ "▁o f",
+ "▁ of",
+ "s t",
+ "▁a nd",
+ "▁an d",
+ "▁ and",
+ "▁ l",
+ "▁t h",
+ "▁ th",
+ "▁ n",
+ "en t",
+ "e nt",
+ "i l",
+ "c t",
+ "r o",
+ "▁r e",
+ "▁ re",
+ "i d",
+ "a m",
+ "▁ I",
+ "a d",
+ "▁ e",
+ "▁ S",
+ "▁ g",
+ "▁ T",
+ "i m",
+ "o t",
+ "a c",
+ "u r",
+ "▁ (",
+ "i g",
+ "▁ =",
+ "o l",
+ "u t",
+ "▁ A",
+ "s e",
+ "▁ u",
+ "v e",
+ "▁ C",
+ "i f",
+ "o w",
+ "▁ y",
+ "c h",
+ "a y",
+ "▁d e",
+ "▁ de",
+ "▁s t",
+ "▁ st",
+ "▁ |",
+ "ve r",
+ "v er",
+ ") ;",
+ "▁ \"",
+ "l y",
+ "▁b e",
+ "▁ be",
+ "* *",
+ "▁i s",
+ "▁ is",
+ "o d",
+ "▁ M",
+ "at ion",
+ "ati on",
+ "atio n",
+ "u l",
+ "▁f or",
+ "▁fo r",
+ "▁ for",
+ "▁o n",
+ "▁ on",
+ "a g",
+ "c e",
+ "te r",
+ "t er",
+ "i r",
+ "t h",
+ "▁ v",
+ "q u",
+ "▁ B",
+ "e m",
+ "▁ P",
+ "▁y ou",
+ "▁yo u",
+ "▁ you",
+ "▁t hat",
+ "▁th at",
+ "▁ that",
+ "u n",
+ "▁ {",
+ "it h",
+ "i th",
+ "r i",
+ "es t",
+ "e st",
+ "a b",
+ "- -",
+ "a p",
+ "▁i t",
+ "▁ it",
+ "▁c on",
+ "▁co n",
+ "▁ con",
+ "at e",
+ "a te",
+ "u s",
+ "▁ H",
+ "u m",
+ "▁ D",
+ "o s",
+ "p e",
+ "▁ -",
+ "▁w h",
+ "▁ wh",
+ "▁a l",
+ "▁ al",
+ "▁a s",
+ "▁ as",
+ "an d",
+ "a nd",
+ "is t",
+ "i st",
+ "▁ L",
+ "▁ W",
+ "▁w ith",
+ "▁ with",
+ "▁a n",
+ "▁ an",
+ "er e",
+ "e re",
+ "▁ *",
+ "▁ R",
+ "▁h e",
+ "▁ he",
+ "▁ F",
+ "o c",
+ "▁w as",
+ "▁wa s",
+ "▁ was",
+ "er s",
+ "e rs",
+ "k e",
+ "ou t",
+ "o ut",
+ "h t",
+ "▁ r",
+ "es s",
+ "e ss",
+ "o p",
+ "re s",
+ "r es",
+ "i e",
+ "▁ E",
+ "▁ \\",
+ "▁T he",
+ "▁Th e",
+ "▁ The",
+ "en d",
+ "e nd",
+ "l d",
+ "▁ N",
+ "or t",
+ "o rt",
+ "▁ G",
+ "/ /",
+ "▁ #",
+ "ou r",
+ "o ur",
+ "t e",
+ "il l",
+ "i ll",
+ "ai n",
+ "a in",
+ "▁s e",
+ "▁ se",
+ "▁ $",
+ "▁p ro",
+ "▁pr o",
+ "▁ pro",
+ "or e",
+ "o re",
+ "▁c om",
+ "▁co m",
+ "▁ com",
+ "am e",
+ "a me",
+ "t r",
+ "▁n e",
+ "▁ ne",
+ "ro m",
+ "r om",
+ "u b",
+ "▁a t",
+ "▁ at",
+ "▁e x",
+ "▁ ex",
+ "an t",
+ "a nt",
+ "u e",
+ "▁o r",
+ "▁ or",
+ "▁ }",
+ "ar t",
+ "a rt",
+ "ct ion",
+ "▁ k",
+ "p t",
+ "n t",
+ "i v",
+ "d e",
+ "▁ O",
+ "p l",
+ "ur n",
+ "u rn",
+ "ig ht",
+ "igh t",
+ "i ght",
+ "al l",
+ "a ll",
+ "▁t his",
+ "▁th is",
+ "▁ this",
+ "se r",
+ "s er",
+ "av e",
+ "a ve",
+ "▁n ot",
+ "▁no t",
+ "▁ not",
+ "▁a re",
+ "▁ar e",
+ "▁ are",
+ "▁ j",
+ "▁l e",
+ "▁ le",
+ "i z",
+ "▁ '",
+ "ag e",
+ "a ge",
+ "me nt",
+ "men t",
+ "m ent",
+ "▁t r",
+ "▁ tr",
+ "ac k",
+ "a ck",
+ "us t",
+ "u st",
+ "( )",
+ "- >",
+ "it y",
+ "i ty",
+ "in e",
+ "i ne",
+ "ou ld",
+ "oul d",
+ "o uld",
+ "▁ J",
+ "o g",
+ "▁f rom",
+ "▁fr om",
+ "▁fro m",
+ "▁ from",
+ "▁w e",
+ "▁ we",
+ "el l",
+ "e ll",
+ "▁s h",
+ "▁ sh",
+ "▁e n",
+ "▁ en",
+ "ur e",
+ "u re",
+ "por t",
+ "po rt",
+ "p ort",
+ "▁c h",
+ "▁ ch",
+ "n e",
+ "▁b y",
+ "▁ by",
+ "pe r",
+ "p er",
+ "ar d",
+ "a rd",
+ "as s",
+ "a ss",
+ "g e",
+ "a k",
+ "ar e",
+ "a re",
+ "o k",
+ "a v",
+ "iv e",
+ "i ve",
+ "f f",
+ "ie s",
+ "i es",
+ "at h",
+ "a th",
+ "tu rn",
+ "t urn",
+ "▁ U",
+ "in t",
+ "i nt",
+ "-- --",
+ "--- -",
+ "- ---",
+ "▁i m",
+ "▁ im",
+ "os t",
+ "o st",
+ "ia l",
+ "i al",
+ "▁h ave",
+ "▁ha ve",
+ "▁hav e",
+ "▁ have",
+ "in d",
+ "i nd",
+ "i p",
+ "an s",
+ "a ns",
+ "x t",
+ "▁d o",
+ "▁ do",
+ "c l",
+ "▁i f",
+ "▁ if",
+ "co n",
+ "c on",
+ "i a",
+ "▁h is",
+ "▁hi s",
+ "▁ his",
+ "ul t",
+ "u lt",
+ "ro u",
+ "r ou",
+ "▁s u",
+ "▁ su",
+ "r a",
+ "▁u n",
+ "▁ un",
+ "ab le",
+ "abl e",
+ "a ble",
+ "▁ <",
+ "▁ K",
+ "om e",
+ "o me",
+ "▁q u",
+ "▁ qu",
+ "ge t",
+ "g et",
+ "▁m e",
+ "▁ me",
+ "as t",
+ "a st",
+ "ec t",
+ "e ct",
+ "▁# #",
+ "▁ ##",
+ "t o",
+ "▁c l",
+ "▁ cl",
+ "▁a b",
+ "▁ ab",
+ "ic e",
+ "i ce",
+ "ir e",
+ "i re",
+ "be r",
+ "b er",
+ "on e",
+ "o ne",
+ "ic h",
+ "i ch",
+ "he n",
+ "h en",
+ "▁c an",
+ "▁ca n",
+ "▁ can",
+ "▁T h",
+ "▁ Th",
+ "▁l a",
+ "▁ la",
+ "▁a ll",
+ "▁al l",
+ "▁ all",
+ "im e",
+ "i me",
+ "il e",
+ "i le",
+ "id e",
+ "i de",
+ "\" ,",
+ "▁p l",
+ "▁ pl",
+ "▁ V",
+ "r u",
+ "or m",
+ "o rm",
+ "▁h ad",
+ "▁ha d",
+ "▁ had",
+ "u d",
+ "as e",
+ "a se",
+ "or d",
+ "o rd",
+ ") ,",
+ "▁h er",
+ "▁he r",
+ "▁ her",
+ "▁I n",
+ "▁ In",
+ "ac e",
+ "a ce",
+ "▁b ut",
+ "▁bu t",
+ "▁ but",
+ "at a",
+ "a ta",
+ ": :",
+ "** **",
+ "*** *",
+ "* ***",
+ "on g",
+ "o ng",
+ "▁ &",
+ ". .",
+ "it e",
+ "i te",
+ "yp e",
+ "y pe",
+ "ac t",
+ "a ct",
+ "od e",
+ "o de",
+ "▁y our",
+ "▁you r",
+ "▁yo ur",
+ "▁ your",
+ "▁o ut",
+ "▁ou t",
+ "▁ out",
+ "▁g o",
+ "▁ go",
+ "li c",
+ "l ic",
+ "al ly",
+ "all y",
+ "▁s o",
+ "▁ so",
+ "or k",
+ "a u",
+ "▁u p",
+ "▁ up",
+ "▁ _",
+ "l l",
+ "= =",
+ "▁m y",
+ "▁ my",
+ "p p",
+ "c c",
+ "▁/ /",
+ "▁ //",
+ "▁the y",
+ "▁th ey",
+ "▁ they",
+ "g h",
+ "▁u s",
+ "▁ us",
+ "i b",
+ "ion s",
+ "io ns",
+ "i ons",
+ "ac h",
+ "a ch",
+ "en s",
+ "e ns",
+ "▁a r",
+ "▁ ar",
+ "o b",
+ "el f",
+ "oo k",
+ "o ok",
+ "at ed",
+ "ate d",
+ "a ted",
+ "an g",
+ "a ng",
+ "ig n",
+ "i gn",
+ "▁re turn",
+ "▁r eturn",
+ "▁ret urn",
+ "▁ return",
+ "▁re s",
+ "▁r es",
+ "▁ res",
+ "c k",
+ "ou s",
+ "o us",
+ "с т",
+ ") .",
+ "▁ п",
+ ". \"",
+ "н а",
+ "▁ i",
+ "ai l",
+ "a il",
+ "e p",
+ "▁a d",
+ "▁ ad",
+ "an ce",
+ "anc e",
+ "( \"",
+ "▁* *",
+ "▁ **",
+ "th er",
+ "the r",
+ "t her",
+ "ak e",
+ "a ke",
+ "▁w ill",
+ "▁ will",
+ "▁c omp",
+ "▁com p",
+ "▁co mp",
+ "▁ comp",
+ "▁o ne",
+ "▁on e",
+ "▁ one",
+ "▁g et",
+ "▁ge t",
+ "▁ get",
+ "o v",
+ "▁ Y",
+ "ar y",
+ "a ry",
+ "oc k",
+ "o ck",
+ "▁s he",
+ "▁sh e",
+ "▁ she",
+ "ch e",
+ "c he",
+ "f t",
+ "▁n ew",
+ "▁ne w",
+ "▁ new",
+ "▁d es",
+ "▁de s",
+ "▁ des",
+ "▁l i",
+ "▁ li",
+ "en ce",
+ "enc e",
+ "▁s a",
+ "▁ sa",
+ "re ss",
+ "res s",
+ "r ess",
+ "▁e l",
+ "▁ el",
+ "▁u nd",
+ "▁un d",
+ "▁ und",
+ "e g",
+ "fe r",
+ "f er",
+ "r y",
+ "ea r",
+ "e ar",
+ "os e",
+ "o se",
+ "ve ry",
+ "ver y",
+ "v ery",
+ "' ,",
+ "▁ +",
+ "▁ в",
+ "▁H e",
+ "▁ He",
+ "ub lic",
+ "ubl ic",
+ "u blic",
+ "▁the ir",
+ "iz e",
+ "i ze",
+ "▁w ere",
+ "▁we re",
+ "▁wer e",
+ "▁ were",
+ "in k",
+ "ow n",
+ "o wn",
+ "I n",
+ "{ \\",
+ "▁h as",
+ "▁ha s",
+ "▁ has",
+ "▁p er",
+ "▁pe r",
+ "▁ per",
+ "▁I t",
+ "▁ It",
+ "▁S t",
+ "▁ St",
+ "he r",
+ "h er",
+ "je ct",
+ "j ect",
+ "р а",
+ "il d",
+ "i ld",
+ "s o",
+ "▁s p",
+ "▁ sp",
+ "н и",
+ "d u",
+ "ro w",
+ "r ow",
+ "al ue",
+ "alu e",
+ "se t",
+ "s et",
+ "fo rm",
+ "for m",
+ "f orm",
+ "co m",
+ "c om",
+ "▁m an",
+ "▁ma n",
+ "▁ man",
+ "on t",
+ "o nt",
+ "ul l",
+ "u ll",
+ "▁c ont",
+ "▁con t",
+ "▁co nt",
+ "▁ cont",
+ "▁m ore",
+ "▁mor e",
+ "▁mo re",
+ "▁ more",
+ "ic k",
+ "i ck",
+ "▁w ould",
+ "▁wo uld",
+ "▁e v",
+ "▁ ev",
+ "▁ab out",
+ "▁ about",
+ "it ion",
+ "iti on",
+ "▁ z",
+ "ou nd",
+ "oun d",
+ "o und",
+ "re e",
+ "r ee",
+ "▁C h",
+ "▁ Ch",
+ "▁wh ich",
+ "▁ which",
+ "i o",
+ "() ;",
+ "( );",
+ "▁w ho",
+ "▁wh o",
+ "▁ who",
+ "er r",
+ "e rr",
+ "or y",
+ "o ry",
+ "ou nt",
+ "oun t",
+ "o unt",
+ "at ions",
+ "ation s",
+ "ati ons",
+ "atio ns",
+ "▁ с",
+ "ri ng",
+ "rin g",
+ "r ing",
+ "< /",
+ "▁f e",
+ "▁ fe",
+ "к о",
+ "н о",
+ "▁d is",
+ "▁di s",
+ "▁ dis",
+ "m a",
+ "▁t hem",
+ "▁the m",
+ "▁th em",
+ "▁a ny",
+ "▁an y",
+ "▁ any",
+ "▁n o",
+ "▁ no",
+ "-- ------",
+ "---- ----",
+ "--- -----",
+ "----- ---",
+ "------ --",
+ "------- -",
+ "- -------",
+ "▁p re",
+ "▁pr e",
+ "▁ pre",
+ "▁t e",
+ "▁ te",
+ "▁r o",
+ "▁ ro",
+ "▁h im",
+ "▁hi m",
+ "▁ him",
+ "▁ :",
+ "u p",
+ "▁in t",
+ "▁i nt",
+ "▁ int",
+ "▁a g",
+ "▁ ag",
+ "S t",
+ "ar k",
+ "e x",
+ "p h",
+ "ie nt",
+ "ien t",
+ "i ent",
+ "el y",
+ "e ly",
+ "▁p r",
+ "▁ pr",
+ "E R",
+ "▁im port",
+ "▁imp ort",
+ "▁ import",
+ "▁t ime",
+ "▁tim e",
+ "▁ti me",
+ "▁ time",
+ "р о",
+ "pr o",
+ "p ro",
+ "Us er",
+ "Use r",
+ "U ser",
+ "l o",
+ "▁ /",
+ "▁ [",
+ "or s",
+ "o rs",
+ "= \"",
+ "▁t here",
+ "▁the re",
+ "▁th ere",
+ "▁ther e",
+ "▁ there",
+ "▁l ike",
+ "▁li ke",
+ "▁lik e",
+ "▁ like",
+ "ol d",
+ "o ld",
+ "▁w hen",
+ "▁wh en",
+ "▁whe n",
+ "▁ when",
+ "ve rs",
+ "ver s",
+ "v ers",
+ "▁s ome",
+ "▁so me",
+ "▁som e",
+ "▁ some",
+ "in gs",
+ "ing s",
+ ") )",
+ "▁p art",
+ "▁par t",
+ "▁pa rt",
+ "▁ part",
+ "ic al",
+ "ica l",
+ "i cal",
+ "▁f un",
+ "▁fu n",
+ "▁ fun",
+ "▁k n",
+ "▁ kn",
+ "ay s",
+ "a ys",
+ "ie r",
+ "i er",
+ "▁b een",
+ "▁be en",
+ "ov e",
+ "o ve",
+ "▁s c",
+ "▁ sc",
+ "ia n",
+ "i an",
+ "▁o ver",
+ "▁ov er",
+ "▁ over",
+ "ie l",
+ "i el",
+ "▁p e",
+ "▁ pe",
+ "ri b",
+ "r ib",
+ "pu t",
+ "p ut",
+ "e c",
+ "et h",
+ "e th",
+ "ar am",
+ "ara m",
+ "a ram",
+ "ap p",
+ "a pp",
+ "▁ –",
+ "▁s tat",
+ "▁st at",
+ "▁sta t",
+ "▁ stat",
+ "po n",
+ "p on",
+ "▁w hat",
+ "▁wh at",
+ "▁ what",
+ "pt ion",
+ "w e",
+ "ad e",
+ "a de",
+ "▁w ork",
+ "▁wor k",
+ "▁ work",
+ "te xt",
+ "tex t",
+ "t ext",
+ "▁s aid",
+ "▁sa id",
+ "▁# ##",
+ "▁## #",
+ "▁ ###",
+ "I N",
+ "▁j ust",
+ "▁ju st",
+ "▁ just",
+ "ir st",
+ "irs t",
+ "▁in to",
+ "▁int o",
+ "▁ into",
+ "▁con st",
+ "▁cons t",
+ "▁ const",
+ "our ce",
+ "t t",
+ "p s",
+ "p r",
+ "er v",
+ "e rv",
+ "it t",
+ "i tt",
+ "u g",
+ "_ {",
+ "en ts",
+ "ent s",
+ "is h",
+ "i sh",
+ "en er",
+ "ene r",
+ "e ner",
+ "▁in ter",
+ "▁int er",
+ "▁inte r",
+ "▁ inter",
+ "pl e",
+ "p le",
+ "ol l",
+ "o ll",
+ "me r",
+ "m er",
+ "at er",
+ "ate r",
+ "a ter",
+ "oo l",
+ "o ol",
+ "e f",
+ "▁p ublic",
+ "▁pub lic",
+ "▁pu blic",
+ "▁publi c",
+ "▁ public",
+ "▁o ther",
+ "▁ot her",
+ "▁ other",
+ "р е",
+ "▁d ef",
+ "▁de f",
+ "▁ def",
+ "▁ @",
+ "г о",
+ "oin t",
+ "oi nt",
+ "o int",
+ "▁o ff",
+ "▁of f",
+ "▁ off",
+ "oi d",
+ "o id",
+ "re turn",
+ "ret urn",
+ "r eturn",
+ "▁s et",
+ "▁se t",
+ "▁ set",
+ "w o",
+ "ft er",
+ "fte r",
+ "f ter",
+ "s h",
+ "** ******",
+ "**** ****",
+ "****** **",
+ "▁o ur",
+ "▁ou r",
+ "▁ our",
+ "ri v",
+ "r iv",
+ "is s",
+ "i ss",
+ "▁W e",
+ "▁ We",
+ "n g",
+ "▁o b",
+ "▁ ob",
+ "s s",
+ "g r",
+ "▁t han",
+ "▁th an",
+ "▁ than",
+ "pe ct",
+ "pec t",
+ "p ect",
+ "ie d",
+ "i ed",
+ "s c",
+ "ie w",
+ "i ew",
+ "de r",
+ "d er",
+ "ys t",
+ "y st",
+ "e v",
+ "▁c ould",
+ "▁co uld",
+ "▁cou ld",
+ "▁ could",
+ "an n",
+ "a nn",
+ "en c",
+ "e nc",
+ "O N",
+ "i x",
+ "an c",
+ "a nc",
+ "▁al so",
+ "▁als o",
+ "▁ also",
+ "re at",
+ "rea t",
+ "▁a m",
+ "▁ am",
+ "▁b ec",
+ "▁be c",
+ "▁ bec",
+ "▁ и",
+ "ua l",
+ "u al",
+ "pe c",
+ "p ec",
+ "▁ .",
+ "▁b l",
+ "▁ bl",
+ "le ct",
+ "l ect",
+ "op le",
+ "opl e",
+ "o ple",
+ "y s",
+ "▁g r",
+ "▁ gr",
+ "ic t",
+ "i ct",
+ "i k",
+ "tr ing",
+ "tri ng",
+ "t ring",
+ "▁T his",
+ "▁Th is",
+ "▁ This",
+ "▁b ack",
+ "▁ba ck",
+ "▁ back",
+ "▁ о",
+ "▁f in",
+ "▁fi n",
+ "▁ fin",
+ "at ch",
+ "Co n",
+ "C on",
+ "( '",
+ "er m",
+ "e rm",
+ "▁= =",
+ "▁ ==",
+ "_ _",
+ "na me",
+ "nam e",
+ "n ame",
+ ", \"",
+ "▁d id",
+ "▁di d",
+ "▁ did",
+ "is e",
+ "i se",
+ "▁on ly",
+ "▁ only",
+ "ru ct",
+ "r uct",
+ "le s",
+ "l es",
+ "▁t hen",
+ "▁the n",
+ "▁th en",
+ "▁ then",
+ "au se",
+ "aus e",
+ "a use",
+ "в а",
+ "▁it s",
+ "▁i ts",
+ "▁ its",
+ "ri t",
+ "r it",
+ "▁k now",
+ "▁kn ow",
+ "▁ know",
+ "ie ld",
+ "iel d",
+ "i eld",
+ "▁c lass",
+ "▁cl ass",
+ "▁clas s",
+ "▁ class",
+ "▁ >",
+ "▁e m",
+ "▁ em",
+ "▁$ \\",
+ "▁ $\\",
+ "▁y ear",
+ "▁ye ar",
+ "▁ year",
+ "w n",
+ "} ,",
+ "▁d el",
+ "▁de l",
+ "▁ del",
+ "al e",
+ "a le",
+ "t y",
+ "fi g",
+ "f ig",
+ "s p",
+ "he d",
+ "h ed",
+ "ro und",
+ "rou nd",
+ "r ound",
+ "e w",
+ "▁d i",
+ "▁ di",
+ "▁d er",
+ "▁de r",
+ "▁ der",
+ "р и",
+ "re d",
+ "r ed",
+ "th is",
+ "t his",
+ "le t",
+ "l et",
+ "R E",
+ "a x",
+ "f r",
+ "ess age",
+ "essa ge",
+ "ou gh",
+ "o ugh",
+ "▁c omm",
+ "▁com m",
+ "▁co mm",
+ "▁ comm",
+ "f o",
+ "uc h",
+ "u ch",
+ "o y",
+ "▁pe ople",
+ "▁ people",
+ "yst em",
+ "ys tem",
+ "▁f irst",
+ "▁fir st",
+ "▁ first",
+ "▁f unction",
+ "▁fun ction",
+ "▁ function",
+ "an ge",
+ "ang e",
+ "▁h ow",
+ "▁ho w",
+ "▁ how",
+ "▁e t",
+ "▁ et",
+ "a h",
+ "▁l ook",
+ "▁lo ok",
+ "▁ look",
+ "т о",
+ "un d",
+ "u nd",
+ "▁u nder",
+ "▁un der",
+ "▁und er",
+ "▁ under",
+ "к а",
+ "▁ !",
+ "ra y",
+ "r ay",
+ "S T",
+ "if ic",
+ "ifi c",
+ "i fic",
+ "л и",
+ "re ad",
+ "rea d",
+ "r ead",
+ "▁b et",
+ "▁be t",
+ "▁ bet",
+ "io us",
+ "i ous",
+ "ar g",
+ "a rg",
+ "▁n eed",
+ "▁ne ed",
+ "▁ need",
+ "ma th",
+ "mat h",
+ "m ath",
+ "▁н а",
+ "▁ на",
+ "er t",
+ "e rt",
+ "▁o p",
+ "▁ op",
+ "▁a cc",
+ "▁ac c",
+ "▁ acc",
+ "Pr o",
+ "P ro",
+ "▁e st",
+ "▁es t",
+ "▁ est",
+ "▁U n",
+ "▁ Un",
+ "▁e nt",
+ "▁en t",
+ "▁ ent",
+ "▁re c",
+ "▁r ec",
+ "▁ rec",
+ "▁u se",
+ "▁us e",
+ "▁ use",
+ "е н",
+ "▁p ar",
+ "▁pa r",
+ "▁ par",
+ "a z",
+ "▁ д",
+ "▁W h",
+ "▁ Wh",
+ "sel f",
+ "s elf",
+ "▁k e",
+ "▁ ke",
+ "т а",
+ "▁w ant",
+ "▁wa nt",
+ "▁ want",
+ "▁e nd",
+ "▁en d",
+ "▁ end",
+ "▁d on",
+ "▁do n",
+ "▁ don",
+ "e k",
+ "re n",
+ "r en",
+ "Na me",
+ "N ame",
+ "▁= >",
+ "▁ =>",
+ "▁a pp",
+ "▁ap p",
+ "▁ app",
+ "▁qu e",
+ "▁q ue",
+ "▁ que",
+ "ig h",
+ "i gh",
+ "▁b u",
+ "▁ bu",
+ "eq u",
+ "e qu",
+ "ve l",
+ "v el",
+ "▁a ct",
+ "▁ac t",
+ "▁ act",
+ "cr e",
+ "c re",
+ "A T",
+ "▁v ar",
+ "▁va r",
+ "▁ var",
+ "ce ss",
+ "ces s",
+ "c ess",
+ "== ==",
+ "=== =",
+ "= ===",
+ "E x",
+ "▁a dd",
+ "▁ad d",
+ "▁ add",
+ "▁m od",
+ "▁mo d",
+ "▁ mod",
+ "un g",
+ "u ng",
+ "▁w here",
+ "▁wh ere",
+ "▁whe re",
+ "▁ where",
+ "ni ng",
+ "n ing",
+ "▁f l",
+ "▁ fl",
+ "al s",
+ "a ls",
+ "ter n",
+ "te rn",
+ "t ern",
+ "} }",
+ "▁A l",
+ "▁ Al",
+ "▁p os",
+ "▁po s",
+ "▁ pos",
+ "an k",
+ "▁a p",
+ "▁ ap",
+ "en g",
+ "e ng",
+ "▁ “",
+ "bl e",
+ "b le",
+ "▁re g",
+ "▁r eg",
+ "▁ reg",
+ "^ {",
+ "▁S he",
+ "▁Sh e",
+ "▁ She",
+ "▁* /",
+ "▁ */",
+ "ud e",
+ "u de",
+ "ad d",
+ "a dd",
+ "▁t wo",
+ "▁tw o",
+ "▁ two",
+ "▁c ol",
+ "▁co l",
+ "▁ col",
+ "▁s m",
+ "▁ sm",
+ "ai r",
+ "a ir",
+ "▁m ay",
+ "▁ma y",
+ "▁ may",
+ "fo re",
+ "for e",
+ "f ore",
+ "▁Y ou",
+ "▁ You",
+ "ro ugh",
+ "rou gh",
+ "r ough",
+ "▁c he",
+ "▁ch e",
+ "▁ che",
+ "▁a tt",
+ "▁at t",
+ "▁ att",
+ "ot h",
+ "o th",
+ "л а",
+ "▁c o",
+ "▁ co",
+ "at es",
+ "ate s",
+ "a tes",
+ "▁re m",
+ "▁r em",
+ "▁ rem",
+ "oo d",
+ "o od",
+ "Ty pe",
+ "Typ e",
+ "T ype",
+ "le d",
+ "l ed",
+ "fu l",
+ "f ul",
+ "▁s elf",
+ "▁sel f",
+ "▁ self",
+ "o f",
+ "▁A r",
+ "▁ Ar",
+ "qu e",
+ "q ue",
+ "▁e very",
+ "▁ev ery",
+ "▁ever y",
+ "▁ every",
+ "re f",
+ "r ef",
+ "Th e",
+ "T he",
+ "▁A nd",
+ "▁An d",
+ "▁ And",
+ "▁re l",
+ "▁r el",
+ "▁ rel",
+ "O R",
+ "I d",
+ "▁e ven",
+ "▁ev en",
+ "▁ even",
+ "E N",
+ "▁h and",
+ "▁ha nd",
+ "▁han d",
+ "▁ hand",
+ "ai t",
+ "a it",
+ "▁sh ould",
+ "▁ should",
+ "▁a fter",
+ "▁af ter",
+ "▁ after",
+ "▁d if",
+ "▁di f",
+ "gh t",
+ "g ht",
+ "if e",
+ "i fe",
+ "at or",
+ "ato r",
+ "a tor",
+ "as h",
+ "a sh",
+ "ri but",
+ "rib ut",
+ "ribu t",
+ "um ber",
+ "umb er",
+ "u mber",
+ "▁s ee",
+ "▁se e",
+ "▁ see",
+ "m s",
+ "▁c all",
+ "▁cal l",
+ "▁ca ll",
+ "▁ call",
+ "y n",
+ "d d",
+ "▁e s",
+ "▁ es",
+ "▁m ake",
+ "▁ma ke",
+ "▁ make",
+ "ot her",
+ "oth er",
+ "othe r",
+ "o ther",
+ "▁ —",
+ "\") ;",
+ "\" );",
+ "st r",
+ "s tr",
+ "▁l ong",
+ "▁lo ng",
+ "▁lon g",
+ "▁ long",
+ "le ment",
+ "lem ent",
+ "l ement",
+ "▁w or",
+ "▁wo r",
+ "▁ wor",
+ "it s",
+ "i ts",
+ "▁I f",
+ "▁ If",
+ "al se",
+ "als e",
+ "л ь",
+ "wa rd",
+ "war d",
+ "w ard",
+ "▁п о",
+ "▁ по",
+ "va l",
+ "v al",
+ "on s",
+ "o ns",
+ "▁ Z",
+ "▁n ow",
+ "▁no w",
+ "▁ now",
+ "da ta",
+ "dat a",
+ "d ata",
+ "am p",
+ "a mp",
+ "en se",
+ "ens e",
+ "▁th rough",
+ "▁thr ough",
+ "▁thro ugh",
+ "▁ through",
+ "▁d own",
+ "▁do wn",
+ "▁dow n",
+ "▁ down",
+ "at t",
+ "a tt",
+ "▁st atic",
+ "▁stat ic",
+ "▁ static",
+ "ic s",
+ "i cs",
+ "# #",
+ "po s",
+ "p os",
+ "▁v oid",
+ "▁vo id",
+ "▁ void",
+ "a w",
+ "ou n",
+ "o un",
+ "▁w ay",
+ "▁wa y",
+ "▁ way",
+ "ib le",
+ "i ble",
+ "ve nt",
+ "ven t",
+ "v ent",
+ "ow er",
+ "owe r",
+ "o wer",
+ "▁th ink",
+ "▁thin k",
+ "▁ think",
+ "t s",
+ "* /",
+ "▁a gain",
+ "▁ag ain",
+ "▁ again",
+ "at ing",
+ "ati ng",
+ "atin g",
+ "a ting",
+ "т е",
+ "ne r",
+ "n er",
+ "▁m ost",
+ "▁mo st",
+ "▁mos t",
+ "▁ most",
+ "li ne",
+ "lin e",
+ "l ine",
+ "y m",
+ "▁s ub",
+ "▁su b",
+ "▁ sub",
+ "er son",
+ "ers on",
+ "▁re qu",
+ "▁r equ",
+ "▁req u",
+ "▁ requ",
+ "A L",
+ "A R",
+ "ab el",
+ "abe l",
+ "a bel",
+ "on d",
+ "o nd",
+ ")) ;",
+ ") );",
+ "▁S e",
+ "▁ Se",
+ "▁B ut",
+ "▁Bu t",
+ "▁ But",
+ "al k",
+ "▁A n",
+ "▁ An",
+ "ne w",
+ "n ew",
+ "▁b ecause",
+ "▁bec ause",
+ "▁ because",
+ "ge r",
+ "g er",
+ "ul ar",
+ "ula r",
+ "u lar",
+ "ro up",
+ "rou p",
+ "r oup",
+ "t a",
+ ".. .",
+ ". ..",
+ "▁c ons",
+ "▁con s",
+ "▁co ns",
+ "▁ cons",
+ "▁r ight",
+ "▁ri ght",
+ "▁rig ht",
+ "▁ right",
+ "▁f r",
+ "▁ fr",
+ "b e",
+ "il y",
+ "i ly",
+ "к и",
+ "▁p h",
+ "▁ ph",
+ "ea d",
+ "e ad",
+ "? \"",
+ "▁g u",
+ "▁ gu",
+ "▁el se",
+ "▁els e",
+ "▁ else",
+ "▁s om",
+ "▁so m",
+ "▁ som",
+ "re nt",
+ "ren t",
+ "r ent",
+ "c o",
+ "em ent",
+ "eme nt",
+ "emen t",
+ "e ment",
+ "▁s tr",
+ "▁st r",
+ "▁ str",
+ "au lt",
+ "aul t",
+ "a ult",
+ "▁ з",
+ "л о",
+ "se rt",
+ "ser t",
+ "s ert",
+ "va r",
+ "v ar",
+ "ty pe",
+ "typ e",
+ "t ype",
+ "▁C om",
+ "▁Co m",
+ "▁ Com",
+ "л е",
+ "in s",
+ "i ns",
+ "m e",
+ "wa y",
+ "w ay",
+ "id ent",
+ "ide nt",
+ "iden t",
+ "▁p rov",
+ "▁pro v",
+ "▁pr ov",
+ "▁ prov",
+ "▁ м",
+ "▁tr ue",
+ "▁ true",
+ "▁P ro",
+ "▁Pr o",
+ "▁ Pro",
+ "f l",
+ "▁s l",
+ "▁ sl",
+ "▁A s",
+ "▁ As",
+ "} \\",
+ "I D",
+ "ue s",
+ "u es",
+ "▁in st",
+ "▁ins t",
+ "▁ inst",
+ "▁n ame",
+ "▁na me",
+ "▁nam e",
+ "▁ name",
+ "o x",
+ "▁ )",
+ "l i",
+ "am es",
+ "ame s",
+ "a mes",
+ "Re s",
+ "R es",
+ "▁s ur",
+ "▁su r",
+ "▁ sur",
+ "par am",
+ "pa ram",
+ "para m",
+ "p aram",
+ "▁st art",
+ "▁star t",
+ "▁sta rt",
+ "▁ start",
+ "a j",
+ "S E",
+ "as k",
+ "a sk",
+ "I T",
+ "St ring",
+ "Str ing",
+ "S tring",
+ "▁a ss",
+ "▁as s",
+ "▁ ass",
+ "▁p lay",
+ "▁pl ay",
+ "▁ play",
+ "ti ng",
+ "t ing",
+ "to n",
+ "t on",
+ "▁b efore",
+ "▁be fore",
+ "▁bef ore",
+ "▁ before",
+ "▁p ol",
+ "▁po l",
+ "▁ pol",
+ "ar ch",
+ "arc h",
+ "▁w ell",
+ "▁we ll",
+ "▁wel l",
+ "▁ well",
+ "Co m",
+ "C om",
+ "an y",
+ "a ny",
+ "ol og",
+ "olo g",
+ "o log",
+ "▁e rr",
+ "▁er r",
+ "▁ err",
+ "▁the se",
+ "▁th ese",
+ "ar s",
+ "a rs",
+ "e b",
+ "▁b r",
+ "▁ br",
+ "▁in cl",
+ "▁inc l",
+ "▁ incl",
+ "▁h el",
+ "▁he l",
+ "▁ hel",
+ "er n",
+ "e rn",
+ "od y",
+ "o dy",
+ "в о",
+ "▁in d",
+ "▁i nd",
+ "▁ ind",
+ "-- --------------",
+ "---- ------------",
+ "-------- --------",
+ "--- -------------",
+ "------------ ----",
+ "----- -----------",
+ "---------- ------",
+ "------ ----------",
+ "------------- ---",
+ "-------------- --",
+ "--------- -------",
+ "------- ---------",
+ "----------- -----",
+ "▁d ata",
+ "▁da ta",
+ "▁dat a",
+ "▁ data",
+ "▁g ood",
+ "▁go od",
+ "▁ good",
+ "L E",
+ "] ,",
+ "▁a v",
+ "▁ av",
+ "▁a c",
+ "▁ ac",
+ "id er",
+ "ide r",
+ "i der",
+ "н е",
+ "▁ Q",
+ "▁m in",
+ "▁mi n",
+ "▁ min",
+ "▁m uch",
+ "▁mu ch",
+ "c i",
+ "el s",
+ "e ls",
+ "▁c ur",
+ "▁cu r",
+ "▁ cur",
+ "▁v alue",
+ "▁val ue",
+ "▁ value",
+ "er y",
+ "e ry",
+ "u f",
+ "▁l oc",
+ "▁lo c",
+ "▁ loc",
+ "re ak",
+ "rea k",
+ "at ive",
+ "ati ve",
+ "ativ e",
+ "im es",
+ "ime s",
+ "i mes",
+ "C l",
+ "▁ ,",
+ "▁s er",
+ "▁se r",
+ "▁ ser",
+ "▁d ie",
+ "▁di e",
+ "▁ die",
+ "▁tr ans",
+ "▁tra ns",
+ "▁ trans",
+ "▁res ult",
+ "▁ result",
+ "ex t",
+ "e xt",
+ "▁a ut",
+ "▁au t",
+ "▁ aut",
+ "la nd",
+ "lan d",
+ "l and",
+ "▁& &",
+ "▁ &&",
+ "C h",
+ "te n",
+ "t en",
+ "} $",
+ "▁t ype",
+ "▁typ e",
+ "▁ty pe",
+ "▁ type",
+ "con d",
+ "co nd",
+ "c ond",
+ "ic es",
+ "ice s",
+ "i ces",
+ "▁v ery",
+ "▁ver y",
+ "▁ve ry",
+ "▁ very",
+ "▁o wn",
+ "▁ own",
+ "▁f il",
+ "▁fi l",
+ "▁ fil",
+ "it ies",
+ "iti es",
+ "i ties",
+ "▁p rodu",
+ "▁pro du",
+ "▁prod u",
+ "▁ produ",
+ "▁re ad",
+ "▁r ead",
+ "▁ read",
+ "▁f orm",
+ "▁for m",
+ "▁fo rm",
+ "▁ form",
+ "▁c ase",
+ "▁cas e",
+ "▁ca se",
+ "▁ case",
+ "at her",
+ "ath er",
+ "a ther",
+ "т и",
+ "д а",
+ "е р",
+ "T h",
+ "au t",
+ "a ut",
+ "▁s pec",
+ "▁sp ec",
+ "▁spe c",
+ "▁ spec",
+ "i j",
+ "b l",
+ "il ity",
+ "ili ty",
+ "▁ é",
+ "▁e r",
+ "▁ er",
+ "▁d oes",
+ "▁do es",
+ "▁ does",
+ "▁h ere",
+ "▁he re",
+ "▁her e",
+ "▁ here",
+ "th e",
+ "t he",
+ "ur es",
+ "ure s",
+ "u res",
+ "▁ %",
+ "mi n",
+ "m in",
+ "▁n ull",
+ "▁nu ll",
+ "▁ null",
+ "ra p",
+ "r ap",
+ "\" )",
+ "r r",
+ "Li st",
+ "L ist",
+ "ri ght",
+ "rig ht",
+ "r ight",
+ "▁U ser",
+ "▁Us er",
+ "▁Use r",
+ "▁ User",
+ "U L",
+ "at ional",
+ "ation al",
+ "ati onal",
+ "atio nal",
+ "▁b eing",
+ "▁be ing",
+ "▁bei ng",
+ "▁ being",
+ "A N",
+ "s k",
+ "▁c ar",
+ "▁ca r",
+ "▁ car",
+ "ol e",
+ "o le",
+ "▁d ist",
+ "▁dis t",
+ "▁di st",
+ "▁ dist",
+ "pl ic",
+ "p lic",
+ "ol low",
+ "oll ow",
+ "▁p res",
+ "▁pre s",
+ "▁pr es",
+ "▁ pres",
+ "▁s uch",
+ "▁su ch",
+ "▁suc h",
+ "▁ such",
+ "re am",
+ "rea m",
+ "in ce",
+ "inc e",
+ "ga n",
+ "g an",
+ "▁F or",
+ "▁Fo r",
+ "▁ For",
+ "\" :",
+ "so n",
+ "s on",
+ "riv ate",
+ "▁y ears",
+ "▁year s",
+ "▁ye ars",
+ "▁s erv",
+ "▁se rv",
+ "▁ser v",
+ "▁ serv",
+ "▁m ade",
+ "▁ma de",
+ "▁mad e",
+ "▁ made",
+ "de f",
+ "d ef",
+ "; \r",
+ "▁g l",
+ "▁ gl",
+ "▁b el",
+ "▁be l",
+ "▁ bel",
+ "▁l ist",
+ "▁li st",
+ "▁ list",
+ "▁c or",
+ "▁co r",
+ "▁ cor",
+ "▁d et",
+ "▁de t",
+ "▁ det",
+ "ce ption",
+ "cept ion",
+ "eg in",
+ "e gin",
+ "▁ б",
+ "▁c har",
+ "▁ch ar",
+ "▁cha r",
+ "▁ char",
+ "tr ans",
+ "tra ns",
+ "▁f am",
+ "▁fa m",
+ "▁! =",
+ "▁ !=",
+ "ou se",
+ "ous e",
+ "o use",
+ "▁d ec",
+ "▁de c",
+ "▁ dec",
+ "ic a",
+ "i ca",
+ "▁m any",
+ "▁man y",
+ "▁ma ny",
+ "▁ many",
+ "ak ing",
+ "aki ng",
+ "a king",
+ "▁ à",
+ "▁s im",
+ "▁si m",
+ "▁ sim",
+ "ag es",
+ "age s",
+ "a ges",
+ "uf f",
+ "u ff",
+ "as ed",
+ "ase d",
+ "a sed",
+ "ma n",
+ "m an",
+ "▁S h",
+ "▁ Sh",
+ "ie t",
+ "i et",
+ "ir ect",
+ "ire ct",
+ "i rect",
+ "▁R e",
+ "▁ Re",
+ "▁d iffer",
+ "▁dif fer",
+ "▁diff er",
+ "▁f ind",
+ "▁fin d",
+ "▁fi nd",
+ "▁ find",
+ "eth od",
+ "▁ \r",
+ "in es",
+ "ine s",
+ "i nes",
+ "▁in v",
+ "▁i nv",
+ "▁ inv",
+ "▁p oint",
+ "▁po int",
+ "▁poi nt",
+ "▁ point",
+ "▁The y",
+ "▁Th ey",
+ "▁ They",
+ "▁u sed",
+ "▁us ed",
+ "▁use d",
+ "▁ used",
+ "ct ions",
+ "ction s",
+ "▁st ill",
+ "i ó",
+ "in ed",
+ "ine d",
+ "i ned",
+ "▁wh ile",
+ "▁ while",
+ "I t",
+ "em ber",
+ "emb er",
+ "e mber",
+ "▁s ay",
+ "▁sa y",
+ "▁ say",
+ "▁he lp",
+ "▁hel p",
+ "▁ help",
+ "▁c re",
+ "▁cr e",
+ "▁ cre",
+ "▁ x",
+ "▁T r",
+ "▁ Tr",
+ "um ent",
+ "ume nt",
+ "umen t",
+ "u ment",
+ "▁s k",
+ "▁ sk",
+ "ou ght",
+ "ough t",
+ "ual ly",
+ "u ally",
+ "m essage",
+ "▁C on",
+ "▁Co n",
+ "▁ Con",
+ "▁m on",
+ "▁mo n",
+ "▁ mon",
+ "ar ed",
+ "are d",
+ "a red",
+ "wor k",
+ "w ork",
+ ") :",
+ "is ter",
+ "ist er",
+ "iste r",
+ "i ster",
+ "ar n",
+ "a rn",
+ "iz ed",
+ "ize d",
+ "i zed",
+ "Dat a",
+ "Da ta",
+ "D ata",
+ "or n",
+ "o rn",
+ "▁h ead",
+ "▁he ad",
+ "▁ head",
+ "D E",
+ "▁L e",
+ "▁ Le",
+ "▁p erson",
+ "▁per son",
+ "▁pers on",
+ "▁ person",
+ "ment s",
+ "men ts",
+ "m ents",
+ "eng th",
+ "e ngth",
+ "▁f alse",
+ "▁fal se",
+ "▁fals e",
+ "▁ false",
+ "▁m ed",
+ "▁me d",
+ "▁ med",
+ "▁D e",
+ "▁ De",
+ "ac he",
+ "ach e",
+ "a che",
+ "it ed",
+ "ite d",
+ "i ted",
+ "▁l et",
+ "▁le t",
+ "▁ let",
+ "▁s how",
+ "▁sh ow",
+ "▁ show",
+ "▁s ame",
+ "▁sa me",
+ "▁sam e",
+ "▁ same",
+ "us s",
+ "u ss",
+ "▁g ener",
+ "▁gen er",
+ "▁ge ner",
+ "▁gene r",
+ "▁ gener",
+ "▁ у",
+ "cu r",
+ "c ur",
+ "▁re al",
+ "▁ real",
+ "ce d",
+ "c ed",
+ "\" >",
+ "st ruct",
+ "str uct",
+ "stru ct",
+ "be gin",
+ "b egin",
+ "ce pt",
+ "cep t",
+ "▁b o",
+ "▁ bo",
+ "ir ed",
+ "ire d",
+ "i red",
+ "▁F r",
+ "▁ Fr",
+ "▁st ud",
+ "▁ stud",
+ "de v",
+ "d ev",
+ "A r",
+ "( \\",
+ "▁C l",
+ "▁ Cl",
+ "we en",
+ "w een",
+ "▁t oo",
+ "▁to o",
+ "▁ too",
+ "▁t est",
+ "▁te st",
+ "▁ test",
+ "▁d ay",
+ "▁da y",
+ "▁ day",
+ "o h",
+ "▁f ollow",
+ "▁fol low",
+ "▁ follow",
+ "at ure",
+ "atur e",
+ "atu re",
+ "z e",
+ "ie n",
+ "i en",
+ "re g",
+ "r eg",
+ "ce s",
+ "c es",
+ "ur ing",
+ "uri ng",
+ "u ring",
+ "am b",
+ "a mb",
+ "in a",
+ "i na",
+ "cr i",
+ "c ri",
+ "▁e d",
+ "▁ ed",
+ "S S",
+ "uc k",
+ "u ck",
+ "▁/ *",
+ "▁ /*",
+ "C T",
+ "▁T here",
+ "▁The re",
+ "▁Th ere",
+ "▁Ther e",
+ "▁ There",
+ "▁t ake",
+ "▁tak e",
+ "▁ta ke",
+ "▁ take",
+ "pa r",
+ "p ar",
+ "ul e",
+ "u le",
+ "ca l",
+ "c al",
+ "fo r",
+ "f or",
+ "** **************",
+ "**** ************",
+ "******** ********",
+ "************ ****",
+ "************** **",
+ "s ource",
+ "▁th ose",
+ "co l",
+ "c ol",
+ "▁e ff",
+ "▁ eff",
+ "mo d",
+ "m od",
+ "con t",
+ "co nt",
+ "c ont",
+ "} {",
+ "▁a round",
+ "▁ar ound",
+ "▁ around",
+ "pr ess",
+ "pre ss",
+ "pres s",
+ "p ress",
+ "b y",
+ "▁go ing",
+ "▁ going",
+ "pon se",
+ "pons e",
+ "▁ С",
+ "▁l ine",
+ "▁li ne",
+ "▁lin e",
+ "▁ line",
+ "da te",
+ "dat e",
+ "d ate",
+ "co de",
+ "cod e",
+ "c ode",
+ "[ '",
+ "▁l ife",
+ "▁li fe",
+ "▁lif e",
+ "▁ life",
+ "as on",
+ "a son",
+ "▁u sing",
+ "▁us ing",
+ "▁ using",
+ "▁v al",
+ "▁va l",
+ "▁ val",
+ "▁d u",
+ "▁ du",
+ "y p",
+ "▁O n",
+ "▁ On",
+ "▁f ound",
+ "▁fo und",
+ "▁fou nd",
+ "▁ found",
+ "ol ut",
+ "olu t",
+ "' ]",
+ "ar ent",
+ "are nt",
+ "aren t",
+ "a rent",
+ "▁s tring",
+ "▁st ring",
+ "▁str ing",
+ "▁stri ng",
+ "▁ string",
+ "▁m et",
+ "▁me t",
+ "▁ met",
+ "▁w r",
+ "▁ wr",
+ "us h",
+ "u sh",
+ "st ring",
+ "str ing",
+ "stri ng",
+ "s tring",
+ "si ze",
+ "s ize",
+ "▁v er",
+ "▁ve r",
+ "▁ ver",
+ "▁e ach",
+ "▁ each",
+ "val ue",
+ "v alue",
+ "▁l ast",
+ "▁la st",
+ "▁las t",
+ "▁ last",
+ "▁g ot",
+ "▁go t",
+ "▁ got",
+ "ve n",
+ "v en",
+ "ba ck",
+ "b ack",
+ "Se t",
+ "S et",
+ "e y",
+ "ro l",
+ "r ol",
+ "▁c r",
+ "▁ cr",
+ "th ing",
+ "t hing",
+ "re t",
+ "r et",
+ "é s",
+ "is m",
+ "i sm",
+ "▁bet ween",
+ "▁ between",
+ "O b",
+ "et hing",
+ "eth ing",
+ "e thing",
+ "m p",
+ "▁l o",
+ "▁ lo",
+ "at s",
+ "a ts",
+ "▁N ew",
+ "▁Ne w",
+ "▁ New",
+ "в и",
+ "ad o",
+ "a do",
+ "de x",
+ "d ex",
+ "д и",
+ "▁p ass",
+ "▁pas s",
+ "▁pa ss",
+ "▁ pass",
+ "w h",
+ "▁d en",
+ "▁de n",
+ "▁ den",
+ "Ge t",
+ "G et",
+ "ap t",
+ "a pt",
+ "▁a sk",
+ "▁as k",
+ "▁ ask",
+ "▁s up",
+ "▁su p",
+ "▁ sup",
+ "Val ue",
+ "V alue",
+ "н ы",
+ "▁t ry",
+ "▁tr y",
+ "▁ try",
+ "lat ion",
+ "l ation",
+ "da y",
+ "d ay",
+ "ne ss",
+ "nes s",
+ "n ess",
+ "et s",
+ "e ts",
+ "▁ex per",
+ "▁exp er",
+ "▁ exper",
+ "T r",
+ "▁M ar",
+ "▁Ma r",
+ "▁ Mar",
+ "se rv",
+ "ser v",
+ "s erv",
+ "b r",
+ "▁n umber",
+ "▁num ber",
+ "▁nu mber",
+ "▁ number",
+ "in al",
+ "ina l",
+ "i nal",
+ "ce nt",
+ "cen t",
+ "c ent",
+ "/ *",
+ "no t",
+ "n ot",
+ "ion al",
+ "io nal",
+ "iona l",
+ "i onal",
+ "▁f inal",
+ "▁fin al",
+ "▁fi nal",
+ "▁ final",
+ "' )",
+ "▁r un",
+ "▁ru n",
+ "▁ run",
+ "ov er",
+ "ove r",
+ "o ver",
+ "▁n ever",
+ "▁ne ver",
+ "▁ never",
+ "u c",
+ "▁h igh",
+ "▁hig h",
+ "▁hi gh",
+ "▁ high",
+ "yl e",
+ "y le",
+ "▁in s",
+ "▁i ns",
+ "▁ ins",
+ "▁b est",
+ "▁be st",
+ "▁bes t",
+ "▁ best",
+ "it tle",
+ "itt le",
+ "ri c",
+ "r ic",
+ "▁s ign",
+ "▁si gn",
+ "▁sig n",
+ "▁ sign",
+ "▁d em",
+ "▁de m",
+ "▁ dem",
+ "in ess",
+ "ine ss",
+ "ines s",
+ "i ness",
+ "g y",
+ "▁w ar",
+ "▁wa r",
+ "▁ war",
+ "is hed",
+ "ish ed",
+ "▁g iv",
+ "▁gi v",
+ "ke y",
+ "k ey",
+ "▁ X",
+ "( $",
+ "▁ch ild",
+ "▁chi ld",
+ "▁ child",
+ "le ss",
+ "les s",
+ "l ess",
+ "way s",
+ "wa ys",
+ "w ays",
+ "in cl",
+ "inc l",
+ "ro p",
+ "r op",
+ "ra w",
+ "r aw",
+ ": //",
+ "▁ «",
+ "n o",
+ "ind ow",
+ "indo w",
+ "f e",
+ "ri end",
+ "rie nd",
+ "rien d",
+ "▁l es",
+ "▁le s",
+ "▁ les",
+ "▁l os",
+ "▁lo s",
+ "▁ los",
+ "fil e",
+ "fi le",
+ "f ile",
+ "form ation",
+ "format ion",
+ "cc ess",
+ "c cess",
+ "▁ В",
+ "n a",
+ "▁i l",
+ "▁ il",
+ "is ion",
+ "isi on",
+ "le r",
+ "l er",
+ "▁a rt",
+ "▁ar t",
+ "▁ art",
+ "Con t",
+ "Co nt",
+ "C ont",
+ "▁w orld",
+ "▁wor ld",
+ "▁ world",
+ "▁t urn",
+ "▁tu rn",
+ "▁tur n",
+ "▁ turn",
+ "▁re ally",
+ "▁real ly",
+ "▁E x",
+ "▁ Ex",
+ "м а",
+ "▁ П",
+ "ter s",
+ "te rs",
+ "t ers",
+ "ar get",
+ "arg et",
+ "arge t",
+ "Er r",
+ "E rr",
+ "▁h app",
+ "▁ha pp",
+ "ti me",
+ "tim e",
+ "t ime",
+ "▁S o",
+ "▁ So",
+ "di v",
+ "d iv",
+ "▁did n",
+ "▁di dn",
+ "ad a",
+ "a da",
+ "oo t",
+ "o ot",
+ "} )",
+ "▁s ch",
+ "▁sc h",
+ "▁ sch",
+ "▁c le",
+ "▁cl e",
+ "▁ cle",
+ "▁some thing",
+ "▁som ething",
+ "▁somet hing",
+ "▁ something",
+ "() .",
+ "( ).",
+ "▁c our",
+ "▁co ur",
+ "▁cou r",
+ "ev er",
+ "eve r",
+ "e ver",
+ "an ts",
+ "ant s",
+ "▁ ?",
+ "T o",
+ "▁ `",
+ "tr y",
+ "t ry",
+ "u x",
+ "ai s",
+ "a is",
+ "ro ss",
+ "ros s",
+ "r oss",
+ "hi p",
+ "h ip",
+ "▁re p",
+ "▁r ep",
+ "▁ rep",
+ "la bel",
+ "lab el",
+ "l abel",
+ "▁b oth",
+ "▁bo th",
+ "▁bot h",
+ "▁ both",
+ "* ,",
+ "ot t",
+ "o tt",
+ "м и",
+ "an e",
+ "a ne",
+ "▁o pen",
+ "▁op en",
+ "▁ open",
+ "w w",
+ "▁c ome",
+ "▁com e",
+ "▁co me",
+ "▁ come",
+ "▁e xt",
+ "▁ex t",
+ "▁ ext",
+ "re m",
+ "r em",
+ "_{ \\",
+ "_ {\\",
+ "▁o ld",
+ "▁ol d",
+ "▁ old",
+ "ch ed",
+ "che d",
+ "c hed",
+ ". _",
+ "M E",
+ "if y",
+ "i fy",
+ "g g",
+ "Co l",
+ "C ol",
+ "vi ew",
+ "v iew",
+ "▁b us",
+ "▁bu s",
+ "▁ bus",
+ "▁m ust",
+ "▁mus t",
+ "▁mu st",
+ "▁ must",
+ "▁d ifferent",
+ "▁differ ent",
+ "lo g",
+ "l og",
+ "is ts",
+ "ist s",
+ "i sts",
+ "ro ll",
+ "rol l",
+ "r oll",
+ "a i",
+ "▁з а",
+ "▁ за",
+ "▁s ystem",
+ "▁sys tem",
+ "▁syst em",
+ "▁ system",
+ "iv ers",
+ "ive rs",
+ "iver s",
+ "i vers",
+ "at us",
+ "atu s",
+ "ot e",
+ "o te",
+ "me d",
+ "m ed",
+ "] .",
+ "ak es",
+ "ake s",
+ "a kes",
+ "R O",
+ "▁c ent",
+ "▁ce nt",
+ "▁ cent",
+ "gr am",
+ "gra m",
+ "g ram",
+ "▁p rivate",
+ "▁priv ate",
+ "▁ private",
+ "▁g reat",
+ "▁gre at",
+ "\" ;",
+ "op y",
+ "o py",
+ "▁fe el",
+ "▁fee l",
+ "▁H ow",
+ "▁Ho w",
+ "▁ How",
+ "// //",
+ "/// /",
+ "/ ///",
+ "I C",
+ "▁d r",
+ "▁ dr",
+ "ain s",
+ "ai ns",
+ "a ins",
+ "lo ck",
+ "loc k",
+ "l ock",
+ "E n",
+ "▁S ch",
+ "▁Sc h",
+ "▁ Sch",
+ "▁m at",
+ "▁ma t",
+ "▁ mat",
+ "▁h ome",
+ "▁hom e",
+ "▁ho me",
+ "▁ home",
+ "per ty",
+ "pert y",
+ "te st",
+ "tes t",
+ "t est",
+ "lo c",
+ "l oc",
+ "▁w om",
+ "▁wo m",
+ "s w",
+ "ar ly",
+ "arl y",
+ "▁E n",
+ "▁ En",
+ "▁к о",
+ "▁ ко",
+ "de n",
+ "d en",
+ "ст а",
+ "с та",
+ "▁ а",
+ "et er",
+ "ete r",
+ "e ter",
+ "▁incl ud",
+ "▁inclu d",
+ "UL L",
+ "U LL",
+ "▁m em",
+ "▁me m",
+ "▁ mem",
+ "▁p o",
+ "▁ po",
+ "▁l ittle",
+ "▁lit tle",
+ "▁litt le",
+ "▁a rg",
+ "▁ar g",
+ "▁ arg",
+ "▁} ,",
+ "▁ },",
+ "in clude",
+ "incl ude",
+ "et a",
+ "e ta",
+ "▁p lace",
+ "▁pl ace",
+ "▁plac e",
+ "▁ place",
+ "id th",
+ "us tom",
+ "ust om",
+ "▁| |",
+ "▁ ||",
+ "▁t em",
+ "▁te m",
+ "▁ tem",
+ "ri ed",
+ "rie d",
+ "r ied",
+ "▁f act",
+ "▁fac t",
+ "▁fa ct",
+ "▁ fact",
+ "ien ce",
+ "i ence",
+ "▁P l",
+ "▁ Pl",
+ "op t",
+ "o pt",
+ "el e",
+ "e le",
+ "g o",
+ "A C",
+ "in ter",
+ "int er",
+ "inte r",
+ "==== ====",
+ "() ,",
+ "( ),",
+ "ot s",
+ "o ts",
+ "ra l",
+ "r al",
+ "iqu e",
+ "iq ue",
+ "i que",
+ "av ing",
+ "avi ng",
+ "a ving",
+ "m l",
+ "▁th ought",
+ "▁though t",
+ "▁thou ght",
+ "fr ac",
+ "f rac",
+ "▁c are",
+ "▁car e",
+ "▁ca re",
+ "▁ care",
+ "() );",
+ "()) ;",
+ "( ));",
+ "▁p ut",
+ "▁pu t",
+ "▁ put",
+ "▁m ight",
+ "▁mi ght",
+ "▁mig ht",
+ "▁A mer",
+ "▁Am er",
+ "▁ Amer",
+ "▁( !",
+ "▁ (!",
+ "am ple",
+ "amp le",
+ "al th",
+ "alt h",
+ "▁f ew",
+ "▁fe w",
+ "▁st ate",
+ "▁stat e",
+ "▁sta te",
+ "▁ state",
+ "su b",
+ "s ub",
+ "▁O r",
+ "▁ Or",
+ "] ;",
+ "▁s ize",
+ "▁si ze",
+ "▁ size",
+ "▁S p",
+ "▁ Sp",
+ "▁with out",
+ "▁ without",
+ "▁p oss",
+ "▁pos s",
+ "▁po ss",
+ "▁ poss",
+ "e q",
+ "pl ay",
+ "p lay",
+ "▁ex pect",
+ "▁exp ect",
+ "▁ expect",
+ "▁se cond",
+ "▁sec ond",
+ "▁ second",
+ "▁S tring",
+ "▁St ring",
+ "▁Str ing",
+ "▁ String",
+ "ui ld",
+ "u ild",
+ "▁n ext",
+ "▁ne xt",
+ "▁ next",
+ "+ +",
+ "re qu",
+ "req u",
+ "r equ",
+ "▁A ll",
+ "▁Al l",
+ "▁ All",
+ "▁m en",
+ "▁me n",
+ "▁ men",
+ "▁W hen",
+ "▁Wh en",
+ "▁Whe n",
+ "▁ When",
+ "it er",
+ "ite r",
+ "i ter",
+ "am ent",
+ "ame nt",
+ "amen t",
+ "a ment",
+ "ne t",
+ "n et",
+ "▁ К",
+ "ro n",
+ "r on",
+ "ain t",
+ "ai nt",
+ "a int",
+ "▁I s",
+ "▁ Is",
+ "в е",
+ "pe nd",
+ "pen d",
+ "p end",
+ "trans lation",
+ "transl ation",
+ "▁г о",
+ "▁ го",
+ "ч е",
+ "▁v an",
+ "▁va n",
+ "▁ van",
+ "▁an other",
+ "▁ano ther",
+ "▁re t",
+ "▁r et",
+ "▁ ret",
+ "▁L a",
+ "▁ La",
+ "Mo d",
+ "M od",
+ "IO N",
+ "I ON",
+ "li st",
+ "l ist",
+ "▁p ost",
+ "▁pos t",
+ "▁po st",
+ "▁ post",
+ "d a",
+ "wa re",
+ "war e",
+ "w are",
+ "▁w ord",
+ "▁wor d",
+ "▁wo rd",
+ "▁ word",
+ "Err or",
+ "Er ror",
+ "▁se em",
+ "▁see m",
+ "▁cont in",
+ "▁ contin",
+ "at ic",
+ "ati c",
+ "▁th ree",
+ "▁thr ee",
+ "▁ three",
+ "Ob ject",
+ "Obj ect",
+ "▁part ic",
+ "▁parti c",
+ "$ .",
+ "▁m ark",
+ "▁mar k",
+ "▁ mark",
+ "▁v is",
+ "▁vi s",
+ "▁ vis",
+ "r c",
+ "▁s w",
+ "▁ sw",
+ "pt ions",
+ "ption s",
+ "▁b reak",
+ "▁bre ak",
+ "▁ break",
+ "▁th ings",
+ "▁thing s",
+ "▁thin gs",
+ "ut e",
+ "u te",
+ "u i",
+ "▁T hat",
+ "▁Th at",
+ "▁ That",
+ "ur s",
+ "u rs",
+ "g l",
+ "р у",
+ "▁f ile",
+ "▁fil e",
+ "▁fi le",
+ "▁ file",
+ "us e",
+ "u se",
+ "ig ned",
+ "ign ed",
+ "igne d",
+ "par t",
+ "pa rt",
+ "p art",
+ "U n",
+ "▁e qu",
+ "▁eq u",
+ "▁ equ",
+ "( &",
+ "▁l ead",
+ "▁le ad",
+ "r m",
+ "ain ed",
+ "ai ned",
+ "aine d",
+ "a ined",
+ "▁B e",
+ "▁ Be",
+ "pat h",
+ "pa th",
+ "p ath",
+ "▁sm all",
+ "▁ small",
+ "ag er",
+ "age r",
+ "a ger",
+ "▁al ways",
+ "▁ always",
+ "▁E l",
+ "▁ El",
+ "▁or der",
+ "▁ord er",
+ "▁ order",
+ "▁e y",
+ "▁ ey",
+ "▁w on",
+ "▁wo n",
+ "▁ won",
+ "ap e",
+ "a pe",
+ "▁l eft",
+ "▁le ft",
+ "▁ left",
+ "av a",
+ "a va",
+ "it em",
+ "ite m",
+ "i tem",
+ "ho r",
+ "h or",
+ "▁a way",
+ "▁aw ay",
+ "▁ away",
+ "b b",
+ "fu n",
+ "f un",
+ "▁I nd",
+ "▁In d",
+ "▁ Ind",
+ "m b",
+ "▁st ruct",
+ "▁str uct",
+ "▁stru ct",
+ "▁ struct",
+ "▁pro cess",
+ "▁proc ess",
+ "▁proces s",
+ "▁ process",
+ "▁s upport",
+ "▁sup port",
+ "▁supp ort",
+ "▁ support",
+ "); \r",
+ ") ;\r",
+ "ió n",
+ "i ón",
+ "L O",
+ "▁o per",
+ "▁op er",
+ "▁ oper",
+ "U T",
+ "▁ ·",
+ "P E",
+ "lo ad",
+ "l oad",
+ "of f",
+ "o ff",
+ "▁N o",
+ "▁ No",
+ "iv es",
+ "ive s",
+ "i ves",
+ "ic an",
+ "ica n",
+ "i can",
+ "▁v e",
+ "▁ ve",
+ "act ion",
+ "a ction",
+ "' ;",
+ "▁v o",
+ "▁ vo",
+ "$ ,",
+ "▁G r",
+ "▁ Gr",
+ "pr e",
+ "p re",
+ "n y",
+ "ain ing",
+ "ai ning",
+ "a ining",
+ "io r",
+ "i or",
+ "in it",
+ "ini t",
+ "i nit",
+ "le ction",
+ "lect ion",
+ "l ection",
+ "ar m",
+ "a rm",
+ "um n",
+ "u mn",
+ "ag s",
+ "a gs",
+ "ц и",
+ "ск о",
+ "с ко",
+ "vers ion",
+ "v ersion",
+ "▁T o",
+ "▁ To",
+ "▁re f",
+ "▁r ef",
+ "▁ ref",
+ "st and",
+ "sta nd",
+ "stan d",
+ "▁A t",
+ "▁ At",
+ "if t",
+ "i ft",
+ "▁e in",
+ "fa ce",
+ "fac e",
+ "f ace",
+ "b o",
+ "if ied",
+ "ifi ed",
+ "ve d",
+ "v ed",
+ "su m",
+ "s um",
+ "un e",
+ "u ne",
+ "it al",
+ "ita l",
+ "i tal",
+ "um p",
+ "u mp",
+ "com m",
+ "co mm",
+ "c omm",
+ "▁m ov",
+ "▁mo v",
+ "▁ mov",
+ "el t",
+ "e lt",
+ "▁v on",
+ "▁vo n",
+ "vel op",
+ "ct or",
+ "c tor",
+ "he ad",
+ "h ead",
+ "cl e",
+ "c le",
+ "▁b uild",
+ "▁bu ild",
+ "▁ build",
+ "in c",
+ "i nc",
+ ". '",
+ "b s",
+ "in fo",
+ "inf o",
+ "ch n",
+ "c hn",
+ "▁we ek",
+ "▁ week",
+ "▁b ook",
+ "▁bo ok",
+ "▁ book",
+ "H E",
+ "ba r",
+ "b ar",
+ "ic ense",
+ "▁W hat",
+ "▁Wh at",
+ "▁ What",
+ "▁qu est",
+ "▁que st",
+ "▁q uest",
+ "▁ quest",
+ "ur ch",
+ "at o",
+ "a to",
+ "le ft",
+ "l eft",
+ "▁m ar",
+ "▁ma r",
+ "▁ mar",
+ "▁t op",
+ "▁to p",
+ "▁ top",
+ "F F",
+ "▁f riend",
+ "▁ friend",
+ "▁b eh",
+ "▁be h",
+ "▁f ield",
+ "▁fi eld",
+ "▁ field",
+ "▁again st",
+ "ra ct",
+ "rac t",
+ "r act",
+ "iz ation",
+ "us er",
+ "use r",
+ "u ser",
+ "ch en",
+ "che n",
+ "c hen",
+ "▁ke ep",
+ "▁ keep",
+ "A D",
+ "it or",
+ "ito r",
+ "i tor",
+ "▁n on",
+ "▁no n",
+ "▁ non",
+ "ir d",
+ "i rd",
+ "op e",
+ "o pe",
+ "▁re st",
+ "▁r est",
+ "▁res t",
+ "▁ rest",
+ "▁d ev",
+ "▁de v",
+ "▁ dev",
+ "▁_ _",
+ "▁ __",
+ "▁u na",
+ "▁un a",
+ "▁ una",
+ "▁t erm",
+ "▁te rm",
+ "▁ter m",
+ "▁ term",
+ "I S",
+ "▁p op",
+ "▁po p",
+ "▁ pop",
+ "ri st",
+ "ris t",
+ "r ist",
+ "▁s ince",
+ "▁sin ce",
+ "▁sinc e",
+ "▁ since",
+ "ve s",
+ "v es",
+ "▁h ard",
+ "▁ha rd",
+ "▁har d",
+ "▁ hard",
+ "p i",
+ "ut il",
+ "uti l",
+ "u til",
+ "▁s oc",
+ "▁so c",
+ "▁ soc",
+ "en e",
+ "e ne",
+ "Ex ception",
+ "▁l ocal",
+ "▁loc al",
+ "▁lo cal",
+ "▁ local",
+ "▁d irect",
+ "▁di rect",
+ "▁dire ct",
+ "▁dir ect",
+ "▁ direct",
+ "▁s ure",
+ "▁su re",
+ "▁sur e",
+ "▁ sure",
+ "▁b ro",
+ "▁br o",
+ "▁ bro",
+ "▁d a",
+ "▁ da",
+ "▁< /",
+ "▁ ",
+ "▁cur rent",
+ "▁curr ent",
+ "▁ current",
+ "' :",
+ "W h",
+ "▁in formation",
+ "▁inform ation",
+ "▁ information",
+ "▁i de",
+ "▁id e",
+ "▁ ide",
+ "▁bet ter",
+ "Te xt",
+ "Tex t",
+ "T ext",
+ "ra ph",
+ "rap h",
+ "r aph",
+ "▁st and",
+ "▁stan d",
+ "▁sta nd",
+ "▁ stand",
+ "▁c heck",
+ "▁che ck",
+ "▁ check",
+ "▁ к",
+ "▁n a",
+ "▁ na",
+ "( (",
+ "ou th",
+ "out h",
+ "o uth",
+ "ap s",
+ "a ps",
+ "▁u nt",
+ "▁un t",
+ "▁ unt",
+ "b f",
+ "▁con f",
+ "▁co nf",
+ "▁ conf",
+ "▁s pe",
+ "▁sp e",
+ "▁ spe",
+ "it le",
+ "i tle",
+ "▁C ol",
+ "▁Co l",
+ "▁ Col",
+ "cl ass",
+ "c lass",
+ "ur al",
+ "ura l",
+ "u ral",
+ "ber s",
+ "be rs",
+ "b ers",
+ "M A",
+ "ess ion",
+ "▁ М",
+ "In fo",
+ "Inf o",
+ "▁B r",
+ "▁ Br",
+ "▁e as",
+ "erv ice",
+ "au s",
+ "a us",
+ "ar i",
+ "a ri",
+ "п о",
+ "▁c oun",
+ "▁co un",
+ "▁cou n",
+ "д е",
+ "() )",
+ "( ))",
+ "li ng",
+ "lin g",
+ "l ing",
+ "E D",
+ "ab ly",
+ "abl y",
+ "▁p at",
+ "▁pa t",
+ "▁ pat",
+ "or g",
+ "o rg",
+ "▁i d",
+ "▁ id",
+ "▁ г",
+ "▁t ell",
+ "▁te ll",
+ "▁tel l",
+ "le x",
+ "l ex",
+ "▁al low",
+ "▁all ow",
+ "▁ allow",
+ "re en",
+ "ree n",
+ "r een",
+ "m y",
+ "▁cons ider",
+ "▁consid er",
+ "▁te am",
+ "▁tea m",
+ "▁ team",
+ "le ase",
+ "ht t",
+ "h tt",
+ "▁P r",
+ "▁ Pr",
+ "/* *",
+ "/ **",
+ "▁s ing",
+ "▁si ng",
+ "▁sin g",
+ "▁ sing",
+ "Re qu",
+ "Req u",
+ "R equ",
+ "R e",
+ "id es",
+ "ide s",
+ "i des",
+ "ch es",
+ "che s",
+ "c hes",
+ "▁ob ject",
+ "▁obj ect",
+ "▁ object",
+ "ial ly",
+ "i ally",
+ "B y",
+ "с я",
+ "id ed",
+ "ide d",
+ "i ded",
+ "▁f ree",
+ "▁fr ee",
+ "▁fre e",
+ "▁ free",
+ "▁pro ble",
+ "▁prob le",
+ "ci te",
+ "cit e",
+ "c ite",
+ "▁) ;",
+ "▁ );",
+ "iss ion",
+ "▁d uring",
+ "▁du ring",
+ "▁dur ing",
+ "▁- -",
+ "▁ --",
+ "it her",
+ "ith er",
+ "i ther",
+ "л я",
+ "▁l eg",
+ "▁le g",
+ "▁ leg",
+ "▁s it",
+ "▁si t",
+ "ic ally",
+ "ical ly",
+ "▁k ey",
+ "▁ke y",
+ "▁ key",
+ "le g",
+ "l eg",
+ "tr a",
+ "t ra",
+ "▁m om",
+ "▁mo m",
+ "▁ex pl",
+ "▁exp l",
+ "▁ expl",
+ "▁de velop",
+ "▁ develop",
+ "▁e vent",
+ "▁ev ent",
+ "▁even t",
+ "▁ event",
+ "▁N ULL",
+ "▁ NULL",
+ "oh n",
+ "o hn",
+ "▁// /",
+ "▁/ //",
+ "▁ ///",
+ "▁bus iness",
+ "▁ business",
+ "ч а",
+ "▁pro f",
+ "▁pr of",
+ "▁ prof",
+ "er ror",
+ "err or",
+ "▁p or",
+ "▁po r",
+ "▁ por",
+ "▁com mun",
+ "▁comm un",
+ "▁ commun",
+ "In d",
+ "I nd",
+ "iu m",
+ "i um",
+ "Te st",
+ "T est",
+ "▁A d",
+ "▁ Ad",
+ "ou ble",
+ "▁s on",
+ "▁so n",
+ "▁ son",
+ "ri te",
+ "rit e",
+ "r ite",
+ "re ady",
+ "read y",
+ "rea dy",
+ "▁{ \r",
+ "▁ {\r",
+ "▁t hing",
+ "▁th ing",
+ "▁thin g",
+ "▁ thing",
+ "н я",
+ "▁P h",
+ "▁ Ph",
+ "pe d",
+ "p ed",
+ "с ь",
+ "iv ed",
+ "ive d",
+ "i ved",
+ "Y ou",
+ "ar l",
+ "a rl",
+ "con st",
+ "cons t",
+ ".. /",
+ ". ./",
+ "S e",
+ "S h",
+ "▁p ower",
+ "▁po wer",
+ "▁pow er",
+ "▁ power",
+ "rib ute",
+ "ribut e",
+ "ribu te",
+ "▁M y",
+ "▁ My",
+ "▁t alk",
+ "▁tal k",
+ "▁ talk",
+ "it ch",
+ "▁c alled",
+ "▁call ed",
+ "▁cal led",
+ "▁ called",
+ "▁c ame",
+ "▁cam e",
+ "▁ca me",
+ "▁be lie",
+ "▁bel ie",
+ "U R",
+ "Ad d",
+ "A dd",
+ "▁R es",
+ "▁Re s",
+ "▁ Res",
+ "as ter",
+ "ast er",
+ "aste r",
+ "a ster",
+ "el la",
+ "ell a",
+ "e lla",
+ "ob al",
+ "oba l",
+ "o bal",
+ "▁u ntil",
+ "▁un til",
+ "▁unt il",
+ "▁ until",
+ "▁h um",
+ "▁ hum",
+ "C O",
+ "at ely",
+ "ate ly",
+ "atel y",
+ "## ##",
+ "### #",
+ "# ###",
+ "pu blic",
+ "pub lic",
+ "p ublic",
+ "[ ]",
+ "▁r oom",
+ "▁ro om",
+ "▁ room",
+ "le n",
+ "l en",
+ "▁f amily",
+ "▁fam ily",
+ "▁famil y",
+ "▁ family",
+ "po r",
+ "p or",
+ "▁pro gram",
+ "▁pr ogram",
+ "▁ program",
+ "▁h ist",
+ "▁his t",
+ "▁hi st",
+ "▁ hist",
+ "▁m us",
+ "▁mu s",
+ "▁ mus",
+ "ar ge",
+ "arg e",
+ "on ey",
+ "one y",
+ "o ney",
+ "I m",
+ "el se",
+ "els e",
+ "ail s",
+ "ai ls",
+ "a ils",
+ "a f",
+ "▁l ove",
+ "▁lo ve",
+ "▁lov e",
+ "▁ love",
+ "ä r",
+ "as es",
+ "ase s",
+ "a ses",
+ "ph a",
+ "p ha",
+ "ou rs",
+ "our s",
+ "o urs",
+ "di s",
+ "d is",
+ "ma p",
+ "m ap",
+ "iv er",
+ "ive r",
+ "i ver",
+ "ö r",
+ "▁B l",
+ "▁ Bl",
+ "at eg",
+ "ate g",
+ "st ate",
+ "stat e",
+ "sta te",
+ "St ate",
+ "Stat e",
+ "er tain",
+ "ert ain",
+ "erta in",
+ "▁e ffect",
+ "▁eff ect",
+ "▁ effect",
+ "pr int",
+ "▁b ig",
+ "▁bi g",
+ "▁ big",
+ "in dex",
+ "ind ex",
+ "inde x",
+ "▁p ub",
+ "▁pu b",
+ "▁ pub",
+ "ve rt",
+ "ver t",
+ "v ert",
+ "er o",
+ "e ro",
+ "m d",
+ "▁m ethod",
+ "▁meth od",
+ "▁ method",
+ "▁g ame",
+ "▁gam e",
+ "▁ga me",
+ "▁ game",
+ "ri es",
+ "rie s",
+ "r ies",
+ "le te",
+ "let e",
+ "l ete",
+ "It em",
+ "I tem",
+ "IN G",
+ "I NG",
+ "re sent",
+ "res ent",
+ "al ity",
+ "ali ty",
+ "pt y",
+ "p ty",
+ "le y",
+ "l ey",
+ "oc ument",
+ "▁b eg",
+ "▁be g",
+ "T R",
+ "} .",
+ "▁sch ool",
+ "▁ school",
+ "he s",
+ "h es",
+ "д о",
+ "▁l ot",
+ "▁lo t",
+ "▁ lot",
+ "▁t ook",
+ "▁to ok",
+ "▁too k",
+ "▁a dv",
+ "▁ad v",
+ "▁ adv",
+ "▁c ap",
+ "▁ca p",
+ "▁ cap",
+ "M P",
+ "un k",
+ "▁l ight",
+ "▁li ght",
+ "▁lig ht",
+ "▁ light",
+ "▁l ater",
+ "▁la ter",
+ "▁late r",
+ "▁lat er",
+ ". ,",
+ "Ke y",
+ "K ey",
+ "it ions",
+ "ition s",
+ "iti ons",
+ "▁en ough",
+ "▁/ **",
+ "▁/* *",
+ "▁ /**",
+ "▁w ent",
+ "▁we nt",
+ "▁wen t",
+ "ã o",
+ "▁th ough",
+ "▁thou gh",
+ "▁ though",
+ "▁g roup",
+ "▁gr oup",
+ "▁gro up",
+ "▁ group",
+ "▁me an",
+ "▁ mean",
+ "ск и",
+ "с ки",
+ "A P",
+ "▁n um",
+ "▁nu m",
+ "▁ num",
+ "▁c ond",
+ "▁con d",
+ "▁co nd",
+ "▁ cond",
+ "н і",
+ "▁g iven",
+ "▁giv en",
+ "▁give n",
+ "▁gi ven",
+ "▁w hy",
+ "▁wh y",
+ "▁ why",
+ "▁re ce",
+ "▁rec e",
+ "▁s ide",
+ "▁si de",
+ "▁sid e",
+ "▁ side",
+ "▁f ar",
+ "▁fa r",
+ "▁ far",
+ "Con text",
+ "Cont ext",
+ "м е",
+ "▁l og",
+ "▁lo g",
+ "▁ log",
+ "Vi ew",
+ "V iew",
+ "▁< <",
+ "▁ <<",
+ "fi l",
+ "f il",
+ "ac es",
+ "ace s",
+ "a ces",
+ "en cy",
+ "enc y",
+ "oa d",
+ "o ad",
+ "er ed",
+ "ere d",
+ "e red",
+ "▁pro duct",
+ "▁produ ct",
+ "▁prod uct",
+ "▁ product",
+ "E T",
+ "▁p aram",
+ "▁par am",
+ "▁para m",
+ "▁pa ram",
+ "▁ param",
+ "▁p rote",
+ "▁pro te",
+ "▁pr ote",
+ "▁prot e",
+ "▁ prote",
+ "te s",
+ "t es",
+ "Tim e",
+ "T ime",
+ "j e",
+ "ol ution",
+ "olut ion",
+ "▁р а",
+ "▁ ра",
+ "▁mon th",
+ "▁mont h",
+ "▁ month",
+ "fer ence",
+ "fe rence",
+ "▁a ppe",
+ "▁app e",
+ "▁ap pe",
+ "▁ appe",
+ "▁f ace",
+ "▁fac e",
+ "▁fa ce",
+ "▁ face",
+ "en ed",
+ "ene d",
+ "e ned",
+ "tr act",
+ "tra ct",
+ "t ract",
+ "▁l ess",
+ "▁le ss",
+ "▁les s",
+ "▁ less",
+ "A S",
+ "é e",
+ "▁g ive",
+ "▁giv e",
+ "▁gi ve",
+ "▁k ind",
+ "▁ki nd",
+ "▁kin d",
+ "▁ kind",
+ "▁c ount",
+ "▁co unt",
+ "▁coun t",
+ "▁cou nt",
+ "▁ count",
+ "co unt",
+ "cou nt",
+ "c ount",
+ "▁s top",
+ "▁st op",
+ "▁sto p",
+ "▁ stop",
+ "▁g over",
+ "▁go ver",
+ "k a",
+ "▁err or",
+ "▁er ror",
+ "▁ error",
+ "en ces",
+ "ence s",
+ "enc es",
+ "▁m il",
+ "▁mi l",
+ "▁ mil",
+ "al f",
+ "yn c",
+ "y nc",
+ "vi ous",
+ "v ious",
+ "h o",
+ "▁n ight",
+ "▁ni ght",
+ "▁ night",
+ "er a",
+ "e ra",
+ "▁п ро",
+ "▁пр о",
+ "▁ про",
+ "▁s ol",
+ "▁so l",
+ "▁ sol",
+ "me n",
+ "m en",
+ "▁w ater",
+ "▁wat er",
+ "▁wa ter",
+ "▁ water",
+ "er ing",
+ "eri ng",
+ "e ring",
+ "▁l im",
+ "▁li m",
+ "▁ lim",
+ "Par am",
+ "P aram",
+ "▁h ouse",
+ "▁hous e",
+ "▁ho use",
+ "▁ house",
+ "▁S ystem",
+ "▁ System",
+ "▁p ay",
+ "▁pa y",
+ "▁ pay",
+ "▁: =",
+ "ur o",
+ "u ro",
+ "oc i",
+ "o ci",
+ "z y",
+ "▁al ready",
+ ", \\",
+ "le ngth",
+ "l ength",
+ "▁s i",
+ "▁ si",
+ "▁inter est",
+ "▁inte rest",
+ "▁ interest",
+ "af f",
+ "a ff",
+ "ct ed",
+ "c ted",
+ "ent ion",
+ "enti on",
+ "▁д о",
+ "▁ до",
+ "um e",
+ "u me",
+ "▁app ro",
+ "▁ap pro",
+ "▁ appro",
+ "br e",
+ "b re",
+ "I G",
+ "▁th row",
+ "▁thr ow",
+ "▁thro w",
+ "▁ throw",
+ "math cal",
+ "ir l",
+ "i rl",
+ "▁p rom",
+ "▁pro m",
+ "▁pr om",
+ "▁ prom",
+ "os s",
+ "o ss",
+ "▁re quest",
+ "▁requ est",
+ "▁req uest",
+ "▁ request",
+ "equ ation",
+ "eq uation",
+ "ol ogy",
+ "olog y",
+ "olo gy",
+ "mi t",
+ "m it",
+ "▁p ack",
+ "▁pa ck",
+ "▁pac k",
+ "▁ pack",
+ "in o",
+ "i no",
+ "ar ray",
+ "arr ay",
+ "z a",
+ "ti l",
+ "t il",
+ "U N",
+ "▁p resent",
+ "▁pre sent",
+ "▁pres ent",
+ "▁ present",
+ "▁or gan",
+ "▁org an",
+ "▁ organ",
+ "Fil e",
+ "Fi le",
+ "F ile",
+ "▁o rig",
+ "▁or ig",
+ "▁ orig",
+ "▁f ull",
+ "▁ful l",
+ "▁fu ll",
+ "▁ full",
+ "is tr",
+ "ist r",
+ "i str",
+ "▁f lo",
+ "▁fl o",
+ "h r",
+ "▁as sert",
+ "▁ass ert",
+ "▁ assert",
+ "ar ds",
+ "ard s",
+ "ur l",
+ "u rl",
+ "en n",
+ "e nn",
+ "s l",
+ "▁ А",
+ "▁c ho",
+ "▁ch o",
+ "▁ cho",
+ "▁l evel",
+ "▁le vel",
+ "▁lev el",
+ "▁ level",
+ "O T",
+ "wo rd",
+ "wor d",
+ "w ord",
+ "▁b ody",
+ "▁bo dy",
+ "▁bod y",
+ "▁ body",
+ "▁u ser",
+ "▁us er",
+ "▁use r",
+ "▁ user",
+ "í a",
+ "Q u",
+ "▁m ain",
+ "▁ma in",
+ "▁mai n",
+ "▁ main",
+ "A B",
+ "pl oy",
+ "plo y",
+ "Ev ent",
+ "Even t",
+ "E vent",
+ "▁s uper",
+ "▁su per",
+ "▁sup er",
+ "▁ super",
+ "ok en",
+ "oke n",
+ "o ken",
+ "▁ Н",
+ "A s",
+ "th ers",
+ "ther s",
+ "the rs",
+ "м о",
+ "к у",
+ "▁d ays",
+ "▁day s",
+ "▁da ys",
+ "▁ days",
+ "▁d one",
+ "▁do ne",
+ "▁don e",
+ "▁ done",
+ "▁v iew",
+ "▁vi ew",
+ "▁vie w",
+ "▁ view",
+ "si de",
+ "sid e",
+ "s ide",
+ "с и",
+ "') ;",
+ "' );",
+ "▁v ol",
+ "▁vo l",
+ "▁ vol",
+ "▁t ot",
+ "▁to t",
+ "▁ tot",
+ "ca se",
+ "cas e",
+ "c ase",
+ "▁a ff",
+ "▁af f",
+ "▁ aff",
+ "Requ est",
+ "Re quest",
+ "Req uest",
+ "▁M an",
+ "▁Ma n",
+ "▁ Man",
+ "\\ \\",
+ "▁J ohn",
+ "▁Jo hn",
+ "▁Joh n",
+ "▁ John",
+ "▁ Б",
+ "or th",
+ "ort h",
+ "▁j e",
+ "▁ je",
+ "▁u ne",
+ "▁un e",
+ "▁ une",
+ "l a",
+ "[ \"",
+ "fi eld",
+ "f ield",
+ "▁U S",
+ "▁ US",
+ "ic o",
+ "i co",
+ "▁per form",
+ "▁perf orm",
+ "▁ perform",
+ "ail able",
+ "Con fig",
+ "Conf ig",
+ "O r",
+ "▁mod el",
+ "▁mo del",
+ "▁mode l",
+ "▁ model",
+ "al es",
+ "ale s",
+ "a les",
+ "▁c reate",
+ "▁cre ate",
+ "▁creat e",
+ "▁ create",
+ "▁a nn",
+ "▁an n",
+ "▁ ann",
+ "an ces",
+ "ance s",
+ "anc es",
+ "I L",
+ "in ation",
+ "▁I m",
+ "▁ Im",
+ "an te",
+ "ant e",
+ "a nte",
+ "an a",
+ "a na",
+ "а н",
+ "▁t old",
+ "▁to ld",
+ "con fig",
+ "conf ig",
+ "\" ]",
+ "me t",
+ "m et",
+ "l t",
+ "▁t ext",
+ "▁te xt",
+ "▁tex t",
+ "▁ text",
+ "▁M ay",
+ "▁Ma y",
+ "▁ May",
+ "▁o rg",
+ "▁or g",
+ "▁ org",
+ "▁p ort",
+ "▁po rt",
+ "▁por t",
+ "▁ port",
+ "P l",
+ "ent ly",
+ "▁d oor",
+ "▁do or",
+ "▁ door",
+ "U S",
+ "▁( *",
+ "▁ (*",
+ "k t",
+ "E S",
+ "ent ial",
+ "enti al",
+ "▁is s",
+ "▁i ss",
+ "▁ iss",
+ "▁in c",
+ "▁i nc",
+ "▁ inc",
+ "No de",
+ "N ode",
+ "iv ely",
+ "ive ly",
+ "ivel y",
+ "▁as ked",
+ "▁ask ed",
+ "ir t",
+ "i rt",
+ "▁T e",
+ "▁ Te",
+ "▁re port",
+ "▁rep ort",
+ "▁repo rt",
+ "▁ report",
+ "▁c hang",
+ "▁ch ang",
+ "▁cha ng",
+ "ст и",
+ "с ти",
+ "▁a long",
+ "▁al ong",
+ "▁ch ange",
+ "▁chang e",
+ "▁ change",
+ "Si ze",
+ "S ize",
+ "▁e ver",
+ "▁ev er",
+ "▁ ever",
+ "▁o cc",
+ "▁oc c",
+ "▁ occ",
+ "ur y",
+ "u ry",
+ "▁m ind",
+ "▁min d",
+ "▁mi nd",
+ "▁ mind",
+ "or der",
+ "ord er",
+ "po int",
+ "p oint",
+ "ст о",
+ "с то",
+ "▁w he",
+ "▁wh e",
+ "▁ whe",
+ "▁import ant",
+ "▁ important",
+ "de s",
+ "d es",
+ "▁N ot",
+ "▁No t",
+ "▁ Not",
+ "▁w rit",
+ "▁wr it",
+ "▁ writ",
+ "▁e yes",
+ "▁ey es",
+ "▁eye s",
+ "▁d esc",
+ "▁de sc",
+ "▁des c",
+ "▁ desc",
+ "mo st",
+ "mos t",
+ "m ost",
+ "k s",
+ "▁b it",
+ "▁bi t",
+ "▁ bit",
+ "▁su ccess",
+ "▁suc cess",
+ "▁succ ess",
+ "▁ success",
+ "т ь",
+ "б о",
+ "co re",
+ "cor e",
+ "c ore",
+ "} (",
+ "▁ar ray",
+ "▁arr ay",
+ "▁ array",
+ "li n",
+ "l in",
+ "li sh",
+ "l ish",
+ "▁follow ing",
+ "Fi eld",
+ "F ield",
+ "id s",
+ "i ds",
+ "hi ng",
+ "hin g",
+ "h ing",
+ "▁c al",
+ "▁ca l",
+ "▁ cal",
+ "I s",
+ "ar ing",
+ "ari ng",
+ "arin g",
+ "a ring",
+ "le v",
+ "l ev",
+ "al t",
+ "a lt",
+ "C H",
+ "▁d é",
+ "al pha",
+ "alph a",
+ "▁f our",
+ "▁fo ur",
+ "▁fou r",
+ "▁ four",
+ "▁l aw",
+ "▁la w",
+ "▁ law",
+ "▁с е",
+ "▁ се",
+ "ir on",
+ "iro n",
+ "i ron",
+ "▁d isc",
+ "▁dis c",
+ "▁di sc",
+ "с е",
+ "ke n",
+ "k en",
+ "no de",
+ "nod e",
+ "n ode",
+ "▁P ar",
+ "▁Pa r",
+ "▁ Par",
+ "▁E ng",
+ "▁En g",
+ "▁ Eng",
+ "▁m ove",
+ "▁mov e",
+ "▁mo ve",
+ "▁ move",
+ "▁L icense",
+ "▁Lic ense",
+ "▁ License",
+ "cu l",
+ "c ul",
+ "ion e",
+ "io ne",
+ "i one",
+ ") $",
+ "▁t w",
+ "▁ tw",
+ "W e",
+ "se l",
+ "s el",
+ "▁W ith",
+ "▁Wi th",
+ "▁ With",
+ "▁on ce",
+ "▁ once",
+ "Serv ice",
+ "S ervice",
+ "bo l",
+ "b ol",
+ "ur ed",
+ "ure d",
+ "u red",
+ "id a",
+ "i da",
+ "▁Q u",
+ "▁ Qu",
+ "▁g row",
+ "▁gr ow",
+ "▁gro w",
+ "▁ grow",
+ "▁c onne",
+ "▁con ne",
+ "▁conn e",
+ "▁ conne",
+ "E X",
+ "▁h tt",
+ "▁ htt",
+ "▁} ;",
+ "▁ };",
+ "▁w alk",
+ "▁wal k",
+ "▁ walk",
+ "▁in it",
+ "▁i nit",
+ "▁ init",
+ "na l",
+ "n al",
+ "en der",
+ "end er",
+ "ende r",
+ "e nder",
+ "cri ption",
+ "cript ion",
+ "mb er",
+ "m ber",
+ "le cted",
+ "lect ed",
+ "p o",
+ "▁n il",
+ "▁ni l",
+ "▁ nil",
+ "▁p rob",
+ "▁pro b",
+ "▁pr ob",
+ "▁ prob",
+ "ч и",
+ "▁S te",
+ "▁St e",
+ "▁ Ste",
+ "is on",
+ "iso n",
+ "i son",
+ "an ds",
+ "and s",
+ "os ed",
+ "ose d",
+ "o sed",
+ "ж е",
+ "▁H is",
+ "▁Hi s",
+ "▁ His",
+ "ü r",
+ "Ma n",
+ "M an",
+ "El ement",
+ "Elem ent",
+ "E lement",
+ "▁a ble",
+ "▁ab le",
+ "▁ able",
+ "In dex",
+ "Ind ex",
+ "se arch",
+ "s earch",
+ "▁m ag",
+ "▁ma g",
+ "▁ mag",
+ "а р",
+ "▁c ourse",
+ "▁cour se",
+ "▁cours e",
+ "▁ course",
+ "▁C ar",
+ "▁Ca r",
+ "▁ Car",
+ "▁e xp",
+ "▁ex p",
+ "▁ exp",
+ "ap h",
+ "a ph",
+ "▁m it",
+ "▁mi t",
+ "▁ mit",
+ "▁does n",
+ "▁def ault",
+ "▁ default",
+ "/ >",
+ "ai m",
+ "a im",
+ "▁s ervice",
+ "▁serv ice",
+ "▁ service",
+ "▁with in",
+ "an gu",
+ "ang u",
+ "▁ Д",
+ "uf fer",
+ "uff er",
+ "A G",
+ "▁D o",
+ "▁ Do",
+ "▁in cre",
+ "▁inc re",
+ "▁under stand",
+ "} ^",
+ "▁look ed",
+ "▁lo oked",
+ "ge n",
+ "g en",
+ "ail ed",
+ "ai led",
+ "a iled",
+ "▁ е",
+ "ay er",
+ "aye r",
+ "a yer",
+ "▁O ne",
+ "▁On e",
+ "▁ One",
+ "▁b as",
+ "▁ba s",
+ "▁ bas",
+ "▁j ob",
+ "▁jo b",
+ "▁ job",
+ "m u",
+ "bu t",
+ "b ut",
+ "el ta",
+ "elt a",
+ "▁Ch rist",
+ "▁Chris t",
+ "▁ Christ",
+ "ur ation",
+ "▁re cord",
+ "▁rec ord",
+ "▁ record",
+ "▁Un ivers",
+ "▁ Univers",
+ "iv id",
+ "ivi d",
+ "i vid",
+ "val id",
+ "▁ Р",
+ "▁h old",
+ "▁hol d",
+ "▁ho ld",
+ "▁ hold",
+ "▁t able",
+ "▁tab le",
+ "▁ta ble",
+ "▁ table",
+ "on es",
+ "one s",
+ "o nes",
+ "lin k",
+ "l ink",
+ "▁G e",
+ "▁ Ge",
+ "▁of fer",
+ "▁off er",
+ "st er",
+ "ste r",
+ "s ter",
+ "For m",
+ "F orm",
+ "= {",
+ "▁н е",
+ "▁ не",
+ "st ance",
+ "stan ce",
+ "▁g overn",
+ "▁go vern",
+ "▁gover n",
+ "▁ govern",
+ "▁te chn",
+ "▁tech n",
+ "▁ techn",
+ "▁p rim",
+ "▁pr im",
+ "▁pri m",
+ "▁ prim",
+ "* .",
+ "ch o",
+ "c ho",
+ "ma x",
+ "m ax",
+ "▁f ore",
+ "▁for e",
+ "▁fo re",
+ "▁ fore",
+ "▁C an",
+ "▁Ca n",
+ "▁ Can",
+ "▁pol it",
+ "▁po lit",
+ "▁ polit",
+ "or ies",
+ "ori es",
+ "orie s",
+ "o ries",
+ "▁t imes",
+ "▁time s",
+ "▁tim es",
+ "▁ti mes",
+ "▁ times",
+ "▁d ans",
+ "▁da ns",
+ "▁dan s",
+ "▁a ir",
+ "▁ai r",
+ "▁ air",
+ "▁any thing",
+ "▁s ever",
+ "▁se ver",
+ "ac y",
+ "a cy",
+ "} _",
+ "H e",
+ "▁l east",
+ "▁le ast",
+ "ip s",
+ "i ps",
+ "EN T",
+ "E NT",
+ "d o",
+ "▁о т",
+ "▁ от",
+ "▁c ost",
+ "▁co st",
+ "▁cos t",
+ "▁ cost",
+ ". ”",
+ "▁child ren",
+ "▁ children",
+ "ab ility",
+ "abil ity",
+ "Bu t",
+ "B ut",
+ "▁p ath",
+ "▁pat h",
+ "▁pa th",
+ "▁ path",
+ "res ult",
+ "ac ter",
+ "act er",
+ "▁e lement",
+ "▁el ement",
+ "▁ele ment",
+ "▁elem ent",
+ "▁ element",
+ "e e",
+ "▁w ait",
+ "▁wa it",
+ "▁ wait",
+ "▁m oney",
+ "▁mon ey",
+ "▁mo ney",
+ "Ma p",
+ "M ap",
+ "t d",
+ "oi n",
+ "o in",
+ "iv ing",
+ "ivi ng",
+ "i ving",
+ "ic ht",
+ "ich t",
+ "i cht",
+ "ic y",
+ "i cy",
+ "sc h",
+ "s ch",
+ "st e",
+ "s te",
+ "д у",
+ "or ed",
+ "ore d",
+ "o red",
+ "ou d",
+ "o ud",
+ "il le",
+ "ill e",
+ "i lle",
+ "is ed",
+ "ise d",
+ "i sed",
+ "pl ication",
+ "plic ation",
+ "▁c ustom",
+ "▁cust om",
+ "▁ custom",
+ "▁h aving",
+ "▁ha ving",
+ "▁hav ing",
+ "pon ent",
+ "po nent",
+ "▁B y",
+ "▁ By",
+ "ul es",
+ "ule s",
+ "u les",
+ "ue d",
+ "u ed",
+ "at ter",
+ "att er",
+ "atte r",
+ "An d",
+ "A nd",
+ "it ive",
+ "iti ve",
+ "De f",
+ "D ef",
+ "▁m oment",
+ "▁mom ent",
+ "▁mo ment",
+ "▁ moment",
+ "at erial",
+ "ate rial",
+ "ater ial",
+ "Cl ass",
+ "C lass",
+ "og raph",
+ "ograp h",
+ "o graph",
+ "ik e",
+ "i ke",
+ "▁l arge",
+ "▁larg e",
+ "▁ large",
+ "▁# ###",
+ "▁## ##",
+ "▁### #",
+ "▁ ####",
+ "▁e ither",
+ "du ct",
+ "duc t",
+ "d uct",
+ "▁T hen",
+ "▁The n",
+ "▁Th en",
+ "▁ Then",
+ "▁G u",
+ "▁ Gu",
+ "ole an",
+ "o lean",
+ "pe rt",
+ "per t",
+ "p ert",
+ "▁G et",
+ "▁Ge t",
+ "▁ Get",
+ "▁A b",
+ "▁ Ab",
+ "▁sh ort",
+ "▁ short",
+ "O n",
+ "im ent",
+ "ime nt",
+ "imen t",
+ "i ment",
+ "▁pro ject",
+ "▁ project",
+ "cri pt",
+ "cr ipt",
+ "c ript",
+ "▁incl uding",
+ "▁includ ing",
+ "▁inclu ding",
+ "▁ including",
+ "ни я",
+ "▁m aking",
+ "▁ma king",
+ "▁ making",
+ "▁some one",
+ "▁F l",
+ "▁ Fl",
+ "▁s at",
+ "▁sa t",
+ "▁ sat",
+ "▁comp any",
+ "▁compan y",
+ "▁ company",
+ "oc us",
+ "p u",
+ "▁G od",
+ "▁Go d",
+ "▁ God",
+ "if ication",
+ "ific ation",
+ "N o",
+ "▁s n",
+ "▁ sn",
+ "an o",
+ "a no",
+ "g a",
+ "▁a u",
+ "▁ au",
+ "▁c ou",
+ "▁co u",
+ "▁ cou",
+ "á s",
+ "en ded",
+ "end ed",
+ "ende d",
+ "т у",
+ "ob er",
+ "obe r",
+ "o ber",
+ "▁n othing",
+ "▁not hing",
+ "▁no thing",
+ "▁n et",
+ "▁ne t",
+ "▁ net",
+ "▁p ot",
+ "▁po t",
+ "▁ pot",
+ "▁t yp",
+ "▁ty p",
+ "▁ typ",
+ "▁it em",
+ "▁i tem",
+ "▁ item",
+ "re w",
+ "r ew",
+ "At t",
+ "A tt",
+ "▁you ng",
+ "▁yo ung",
+ "} \r",
+ "nd er",
+ "nde r",
+ "n der",
+ "st art",
+ "sta rt",
+ "star t",
+ "▁S c",
+ "▁ Sc",
+ "* )",
+ "▁e nc",
+ "▁en c",
+ "▁ enc",
+ "▁w omen",
+ "▁wom en",
+ "▁wo men",
+ "▁look ing",
+ "▁lo oking",
+ "▁ looking",
+ "▁р о",
+ "▁ ро",
+ "▁he alth",
+ "▁heal th",
+ "▁ health",
+ "Pat h",
+ "P ath",
+ "▁A fter",
+ "▁Af ter",
+ "▁ After",
+ "▁m ult",
+ "▁mu lt",
+ "▁mul t",
+ "▁ mult",
+ "▁{ \\",
+ "▁ {\\",
+ "▁l and",
+ "▁la nd",
+ "▁lan d",
+ "▁ land",
+ "or ld",
+ "▁D es",
+ "▁De s",
+ "▁ Des",
+ "▁e ng",
+ "▁en g",
+ "▁ eng",
+ "in put",
+ "▁P ol",
+ "▁Po l",
+ "▁ Pol",
+ "\" \"",
+ "Co de",
+ "C ode",
+ "▁s upp",
+ "▁su pp",
+ "▁sup p",
+ "▁ supp",
+ "ain er",
+ "ai ner",
+ "aine r",
+ "a iner",
+ "he ck",
+ "▁m or",
+ "▁mo r",
+ "▁ mor",
+ "▁m ill",
+ "▁mil l",
+ "▁mi ll",
+ "▁ mill",
+ "▁a w",
+ "▁ aw",
+ "f s",
+ "▁do ing",
+ "ting s",
+ "t ings",
+ "ad es",
+ "ade s",
+ "a des",
+ "▁to get",
+ "▁c ertain",
+ "▁cert ain",
+ "▁cer tain",
+ "▁t ogether",
+ "▁toget her",
+ "C E",
+ "ide o",
+ "▁Amer ican",
+ "▁America n",
+ "▁ American",
+ "on y",
+ "o ny",
+ "id d",
+ "i dd",
+ "I I",
+ "ge d",
+ "g ed",
+ "ab les",
+ "able s",
+ "abl es",
+ "a bles",
+ "▁ide nt",
+ "▁id ent",
+ "▁ ident",
+ "io d",
+ "i od",
+ "▁p arent",
+ "▁par ent",
+ "▁pa rent",
+ "▁pare nt",
+ "▁ parent",
+ "F or",
+ "amb da",
+ "an do",
+ "and o",
+ "= \\",
+ "ag ed",
+ "age d",
+ "a ged",
+ "en ding",
+ "end ing",
+ "In t",
+ "I nt",
+ "▁poss ible",
+ "▁ possible",
+ "▁с о",
+ "▁ со",
+ "iv ity",
+ "ivi ty",
+ "nu m",
+ "n um",
+ "r t",
+ "aj or",
+ "ajo r",
+ "a jor",
+ "cre ate",
+ "creat e",
+ "c reate",
+ "ri de",
+ "rid e",
+ "r ide",
+ "▁k new",
+ "▁kn ew",
+ "▁kne w",
+ "bi t",
+ "b it",
+ "it ional",
+ "ition al",
+ "iti onal",
+ "▁l ik",
+ "▁li k",
+ "▁ lik",
+ "▁H er",
+ "▁He r",
+ "▁ Her",
+ "ens ion",
+ "\" .",
+ "ot o",
+ "o to",
+ "▁ex ist",
+ "▁ exist",
+ "ak en",
+ "ake n",
+ "a ken",
+ "▁act ually",
+ "▁actual ly",
+ "c a",
+ "▁ Г",
+ "х о",
+ "in n",
+ "i nn",
+ "Al l",
+ "A ll",
+ "bu f",
+ "b uf",
+ "▁M e",
+ "▁ Me",
+ "▁s een",
+ "▁se en",
+ "▁see n",
+ "▁ seen",
+ "op s",
+ "o ps",
+ "No t",
+ "N ot",
+ "▁cont rol",
+ "▁contr ol",
+ "▁contro l",
+ "▁ control",
+ "▁res pon",
+ "▁resp on",
+ "▁ respon",
+ "} ;",
+ "il t",
+ "i lt",
+ "is k",
+ "i sk",
+ "▁b ad",
+ "▁ba d",
+ "▁ bad",
+ "▁o ften",
+ "▁of ten",
+ "▁p ast",
+ "▁pas t",
+ "▁pa st",
+ "ap er",
+ "ape r",
+ "a per",
+ "▁re ason",
+ "▁ reason",
+ "et ers",
+ "eter s",
+ "ete rs",
+ "e ters",
+ "▁w anted",
+ "▁want ed",
+ "ur a",
+ "u ra",
+ "ta ble",
+ "tab le",
+ "t able",
+ "or mal",
+ "orm al",
+ "wid th",
+ "w idth",
+ "г а",
+ "pt r",
+ "p tr",
+ "▁d est",
+ "▁de st",
+ "▁des t",
+ "▁ dest",
+ "▁de sign",
+ "▁des ign",
+ "▁ design",
+ "▁s ound",
+ "▁so und",
+ "▁sou nd",
+ "▁ sound",
+ "▁p lan",
+ "▁pl an",
+ "▁ plan",
+ "▁b ase",
+ "▁bas e",
+ "▁ba se",
+ "▁ base",
+ "ha nd",
+ "han d",
+ "h and",
+ "g s",
+ "▁s ays",
+ "▁sa ys",
+ "▁say s",
+ "fun ction",
+ "f unction",
+ "▁t ri",
+ "▁tr i",
+ "▁ tri",
+ "m t",
+ "▁in vest",
+ "▁inv est",
+ "▁av ailable",
+ "▁ available",
+ "ay out",
+ "a yout",
+ "▁o ch",
+ "▁oc h",
+ "▁ och",
+ "▁l as",
+ "▁la s",
+ "▁ las",
+ "il led",
+ "ill ed",
+ "ille d",
+ "V al",
+ "▁ ф",
+ "ie ty",
+ "iet y",
+ "i ety",
+ "mo n",
+ "m on",
+ "Ha nd",
+ "H and",
+ "F r",
+ "ia m",
+ "i am",
+ "pa ce",
+ "p ace",
+ "▁O b",
+ "▁ Ob",
+ "▁p ara",
+ "▁par a",
+ "▁pa ra",
+ "▁ para",
+ "▁me et",
+ "▁s um",
+ "▁su m",
+ "▁ sum",
+ "M essage",
+ "ic i",
+ "i ci",
+ "▁k nown",
+ "▁kn own",
+ "▁know n",
+ "▁ known",
+ "▁g en",
+ "▁ge n",
+ "▁ gen",
+ "am ma",
+ "amm a",
+ "a mma",
+ "ar r",
+ "a rr",
+ "▁t re",
+ "▁tr e",
+ "▁ tre",
+ "ok e",
+ "o ke",
+ "ut h",
+ "u th",
+ "~ \\",
+ "▁exper ience",
+ "▁experi ence",
+ "ic le",
+ "icl e",
+ "i cle",
+ "▁I l",
+ "▁ Il",
+ "▁s ent",
+ "▁se nt",
+ "▁sen t",
+ "▁ sent",
+ "▁o thers",
+ "▁other s",
+ "▁ others",
+ "▁s oft",
+ "▁so ft",
+ "▁ soft",
+ "I P",
+ "▁m ax",
+ "▁ma x",
+ "▁ max",
+ "ba ll",
+ "bal l",
+ "b all",
+ "▁mark et",
+ "▁mar ket",
+ "▁ market",
+ "▁p our",
+ "▁po ur",
+ "▁pou r",
+ "pr ession",
+ "press ion",
+ "p ression",
+ "ep s",
+ "e ps",
+ "▁s aw",
+ "▁sa w",
+ "▁a cross",
+ "▁ac ross",
+ "▁S u",
+ "▁ Su",
+ "O ver",
+ "ни е",
+ "ul ation",
+ "u lation",
+ "▁R eg",
+ "▁Re g",
+ "▁ Reg",
+ "▁+ =",
+ "▁ +=",
+ "bo dy",
+ "b ody",
+ ") \\",
+ "▁pr int",
+ "▁pri nt",
+ "▁prin t",
+ "▁ print",
+ "▁п ри",
+ "▁пр и",
+ "▁ при",
+ "d b",
+ "our ces",
+ "ource s",
+ "ward s",
+ "war ds",
+ "w ards",
+ "▁bl ack",
+ "▁ black",
+ "с о",
+ "il i",
+ "i li",
+ "▁E d",
+ "▁ Ed",
+ "▁com plet",
+ "▁comp let",
+ "▁compl et",
+ "▁s ingle",
+ "▁sing le",
+ "▁sin gle",
+ "▁ single",
+ "▁I N",
+ "▁ IN",
+ "ac hed",
+ "ach ed",
+ "ache d",
+ "a ched",
+ "b t",
+ "▁c ode",
+ "▁co de",
+ "▁cod e",
+ "▁ code",
+ "▁b ool",
+ "▁bo ol",
+ "▁ bool",
+ "▁a rea",
+ "▁are a",
+ "▁ar ea",
+ "▁ area",
+ "▁re quire",
+ "▁requ ire",
+ "▁ require",
+ "▁pro blem",
+ "▁proble m",
+ "▁prob lem",
+ "ac ed",
+ "ace d",
+ "a ced",
+ "Eq u",
+ "E qu",
+ "▁con fig",
+ "▁conf ig",
+ "▁ config",
+ "ve c",
+ "v ec",
+ "ne y",
+ "n ey",
+ "c y",
+ "A l",
+ "▁acc ount",
+ "▁ac count",
+ "▁ account",
+ "ym bol",
+ "▁s te",
+ "▁st e",
+ "▁ ste",
+ "ge s",
+ "g es",
+ "Ar ray",
+ "Arr ay",
+ "em pl",
+ "emp l",
+ "con text",
+ "cont ext",
+ "De s",
+ "D es",
+ "Res ult",
+ "ec ut",
+ "e cut",
+ "▁t arget",
+ "▁tar get",
+ "▁ target",
+ "▁get ting",
+ "\" />",
+ "og le",
+ "o gle",
+ "▁him self",
+ "▁was n",
+ "▁wa sn",
+ "▁b lock",
+ "▁bl ock",
+ "▁blo ck",
+ "▁ block",
+ "▁a nt",
+ "▁an t",
+ "▁ ant",
+ "▁Y ork",
+ "▁be come",
+ "▁bec ome",
+ "if f",
+ "i ff",
+ "port s",
+ "por ts",
+ "p orts",
+ "re ate",
+ "reat e",
+ "rea te",
+ "= '",
+ "c d",
+ "loc ation",
+ "l ocation",
+ "е т",
+ "▁a ccess",
+ "▁acc ess",
+ "▁ac cess",
+ "▁ access",
+ "gr ess",
+ "gre ss",
+ "gres s",
+ "g ress",
+ "ro s",
+ "r os",
+ "U p",
+ "▁work ing",
+ "▁wor king",
+ "▁ working",
+ "▁A m",
+ "▁ Am",
+ "iq u",
+ "i qu",
+ "ce r",
+ "c er",
+ "▁( (",
+ "▁ ((",
+ "▁P er",
+ "▁Pe r",
+ "▁ Per",
+ "▁f unc",
+ "▁fun c",
+ "▁fu nc",
+ "▁ func",
+ "▁g irl",
+ "▁gi rl",
+ "▁gir l",
+ "▁ girl",
+ "▁ab ove",
+ "pe n",
+ "p en",
+ "п и",
+ "id o",
+ "i do",
+ "▁v ersion",
+ "▁vers ion",
+ "▁ version",
+ "T Y",
+ "▁ ;",
+ "ma ry",
+ "mar y",
+ "m ary",
+ "ab led",
+ "able d",
+ "abl ed",
+ "a bled",
+ "an nel",
+ "ann el",
+ "anne l",
+ "▁ex ample",
+ "▁exam ple",
+ "▁ example",
+ "▁con text",
+ "▁cont ext",
+ "▁ context",
+ "O P",
+ "▁re d",
+ "▁r ed",
+ "▁ red",
+ "▁c ir",
+ "▁ci r",
+ "▁ cir",
+ "s m",
+ "Lo g",
+ "L og",
+ "▁s pace",
+ "▁sp ace",
+ "▁ space",
+ "▁f ut",
+ "▁fu t",
+ "▁G ener",
+ "▁Ge ner",
+ "▁Gen er",
+ "▁Gene r",
+ "▁ Gener",
+ "il ls",
+ "ill s",
+ "▁d ri",
+ "▁dr i",
+ "_ .",
+ "▁f elt",
+ "▁fe lt",
+ "▁fel t",
+ "▁o ffic",
+ "▁of fic",
+ "▁off ic",
+ "▁= ==",
+ "▁== =",
+ "▁ ===",
+ "i i",
+ "▁start ed",
+ "▁star ted",
+ "▁ Т",
+ "▁} );",
+ "▁}) ;",
+ "▁ });",
+ "j s",
+ "▁fr ont",
+ "▁fro nt",
+ "▁ front",
+ "▁al most",
+ "ir m",
+ "i rm",
+ "! \"",
+ "sign ed",
+ "sig ned",
+ "s igned",
+ "▁y et",
+ "▁ye t",
+ "▁t rad",
+ "▁tr ad",
+ "▁tra d",
+ "ient s",
+ "ien ts",
+ "i ents",
+ "am a",
+ "a ma",
+ "▁in put",
+ "▁ input",
+ "li m",
+ "l im",
+ "п а",
+ "▁к а",
+ "▁ ка",
+ "▁c amp",
+ "▁cam p",
+ "▁ca mp",
+ "▁ camp",
+ "ib r",
+ "i br",
+ "fe ct",
+ "f ect",
+ "un t",
+ "u nt",
+ "▁h alf",
+ "▁hal f",
+ "▁ half",
+ "▁c over",
+ "▁co ver",
+ "▁cov er",
+ "▁ cover",
+ "angu age",
+ "▁b en",
+ "▁be n",
+ "▁ ben",
+ "h a",
+ "▁d iff",
+ "▁di ff",
+ "▁dif f",
+ "▁ diff",
+ "_ \\",
+ "▁о б",
+ "▁ об",
+ "] )",
+ "od es",
+ "ode s",
+ "o des",
+ "he l",
+ "h el",
+ "io s",
+ "i os",
+ "▁ О",
+ "▁m ot",
+ "▁mo t",
+ "▁ mot",
+ "▁s ocial",
+ "▁so cial",
+ "▁soc ial",
+ "▁soci al",
+ "▁ social",
+ "//// ////",
+ "▁s tre",
+ "▁st re",
+ "▁str e",
+ "▁ stre",
+ "gr ound",
+ "gro und",
+ "g round",
+ "і в",
+ "ob ject",
+ "obj ect",
+ "pl es",
+ "ple s",
+ "p les",
+ "re ed",
+ "ree d",
+ "r eed",
+ "▁e en",
+ "▁ een",
+ "▁b ased",
+ "▁bas ed",
+ "▁base d",
+ "▁ba sed",
+ "▁ based",
+ "▁r ange",
+ "▁ran ge",
+ "▁rang e",
+ "▁ range",
+ "A n",
+ "ur g",
+ "u rg",
+ "▁le arn",
+ "▁lear n",
+ "▁ learn",
+ "▁e xc",
+ "▁ex c",
+ "▁ exc",
+ "▁im p",
+ "▁i mp",
+ "▁ imp",
+ "▁me ans",
+ "▁mean s",
+ "▁w ur",
+ "en ds",
+ "end s",
+ "vo id",
+ "v oid",
+ "▁s td",
+ "▁st d",
+ "▁ std",
+ "▁part icular",
+ "▁partic ular",
+ "▁particul ar",
+ "▁parti cular",
+ "j a",
+ "▁s ource",
+ "▁sour ce",
+ "▁ source",
+ "def ault",
+ "p y",
+ "▁a ls",
+ "▁al s",
+ "▁ als",
+ "sc ri",
+ "scr i",
+ "s cri",
+ "st atus",
+ "stat us",
+ "▁st ory",
+ "▁stor y",
+ "▁sto ry",
+ "▁ story",
+ "▁b egin",
+ "▁be gin",
+ "▁beg in",
+ "▁ begin",
+ "▁pos ition",
+ "▁posit ion",
+ "▁ position",
+ "▁spec ial",
+ "▁spe cial",
+ "▁ special",
+ "ph p",
+ "p hp",
+ "▁b ar",
+ "▁ba r",
+ "▁ bar",
+ "▁p ract",
+ "▁pr act",
+ "▁pra ct",
+ "▁prac t",
+ "cal l",
+ "ca ll",
+ "c all",
+ "▁d as",
+ "▁da s",
+ "▁ das",
+ "▁r ad",
+ "▁ra d",
+ "▁ rad",
+ "▁cl ose",
+ "▁clos e",
+ "▁clo se",
+ "▁ close",
+ "ww w",
+ "w ww",
+ "ер е",
+ "е ре",
+ "g u",
+ "▁E r",
+ "▁ Er",
+ "▁d om",
+ "▁do m",
+ "▁ dom",
+ "A M",
+ "▁b ed",
+ "▁be d",
+ "▁ bed",
+ "▁sever al",
+ "au l",
+ "a ul",
+ "bo x",
+ "b ox",
+ "▁l ow",
+ "▁lo w",
+ "▁ low",
+ "pa ck",
+ "p ack",
+ "Re g",
+ "R eg",
+ "O f",
+ "at ures",
+ "ature s",
+ "atur es",
+ "atu res",
+ "é n",
+ "ed er",
+ "ede r",
+ "e der",
+ "uild er",
+ "ca st",
+ "cas t",
+ "c ast",
+ "con om",
+ "co nom",
+ "c onom",
+ "ra ft",
+ "raf t",
+ "r aft",
+ "▁m akes",
+ "▁make s",
+ "▁ma kes",
+ "Lo c",
+ "L oc",
+ "ht tp",
+ "htt p",
+ "h ttp",
+ "▁a bs",
+ "▁ab s",
+ "▁ abs",
+ "re sh",
+ "res h",
+ "r esh",
+ "▁W ill",
+ "▁Wil l",
+ "▁Wi ll",
+ "▁ Will",
+ "bre ak",
+ "b reak",
+ "▁o ptions",
+ "▁opt ions",
+ "▁option s",
+ "▁ options",
+ "fo rt",
+ "for t",
+ "f ort",
+ "▁и з",
+ "▁ из",
+ "▁a nal",
+ "▁an al",
+ "▁ anal",
+ "▁e nv",
+ "▁en v",
+ "▁ env",
+ "( {",
+ "ev ent",
+ "even t",
+ "eve nt",
+ "e vent",
+ "▁p age",
+ "▁pa ge",
+ "▁pag e",
+ "▁ page",
+ "ter nal",
+ "tern al",
+ "▁d istribut",
+ "▁dist ribut",
+ "▁f ood",
+ "▁fo od",
+ "▁foo d",
+ "▁ food",
+ "che ck",
+ "c heck",
+ "C K",
+ "▁в о",
+ "▁ во",
+ "as sert",
+ "ass ert",
+ "asse rt",
+ "á n",
+ "ba se",
+ "bas e",
+ "b ase",
+ "▁w hole",
+ "▁wh ole",
+ "▁who le",
+ "ac ión",
+ "ació n",
+ "aci ón",
+ "a ción",
+ "O D",
+ "▁turn ed",
+ "▁tur ned",
+ "ig ma",
+ "▁res ponse",
+ "▁respon se",
+ "▁respons e",
+ "▁ response",
+ "▁Univers ity",
+ "▁d iv",
+ "▁di v",
+ "▁ div",
+ "ap ter",
+ "apt er",
+ "▁result s",
+ "▁ results",
+ "▁re present",
+ "▁rep resent",
+ "▁every thing",
+ "▁C ent",
+ "▁Ce nt",
+ "▁ Cent",
+ "ut es",
+ "ute s",
+ "u tes",
+ "ri x",
+ "r ix",
+ "▁S ome",
+ "▁So me",
+ "▁Som e",
+ "▁ Some",
+ "▁be hind",
+ "▁beh ind",
+ "▁c reat",
+ "▁cre at",
+ "▁ creat",
+ "pl ace",
+ "plac e",
+ "p lace",
+ "s u",
+ "▁P art",
+ "▁Par t",
+ "▁Pa rt",
+ "▁ Part",
+ "um b",
+ "u mb",
+ "math bb",
+ "pi ng",
+ "pin g",
+ "p ing",
+ "▁m atch",
+ "▁mat ch",
+ "▁ match",
+ "O ut",
+ "do m",
+ "d om",
+ "▁s itu",
+ "▁sit u",
+ "▁si tu",
+ "d r",
+ "ar a",
+ "a ra",
+ "▁w indow",
+ "▁wind ow",
+ "▁ window",
+ "n s",
+ "lish ed",
+ "l ished",
+ "▁V er",
+ "▁Ve r",
+ "▁ Ver",
+ "▁m essage",
+ "▁mess age",
+ "▁ message",
+ "▁E m",
+ "▁ Em",
+ "▁h uman",
+ "▁hum an",
+ "▁ human",
+ "per ties",
+ "pert ies",
+ "л у",
+ "le m",
+ "l em",
+ "OR T",
+ "O RT",
+ "▁e arly",
+ "▁ear ly",
+ "▁qu ick",
+ "▁qui ck",
+ "▁ quick",
+ "▁т а",
+ "▁ та",
+ "ro id",
+ "r oid",
+ "▁c ountry",
+ "▁coun try",
+ "▁count ry",
+ "▁countr y",
+ "▁ country",
+ "▁d ue",
+ "▁du e",
+ "▁ due",
+ "▁D ie",
+ "▁Di e",
+ "▁ Die",
+ "▁t rying",
+ "▁tr ying",
+ "▁try ing",
+ "▁l ive",
+ "▁li ve",
+ "▁liv e",
+ "▁ live",
+ "▁p ress",
+ "▁pre ss",
+ "▁pr ess",
+ "▁pres s",
+ "▁ press",
+ "IN T",
+ "I NT",
+ "W ith",
+ "ov ed",
+ "ove d",
+ "o ved",
+ "▁spec ific",
+ "▁ specific",
+ "▁f all",
+ "▁fa ll",
+ "▁fal l",
+ "▁ fall",
+ "u k",
+ "y l",
+ "▁gener al",
+ "▁gen eral",
+ "▁gene ral",
+ "▁ general",
+ "м у",
+ "н у",
+ "▁n ames",
+ "▁name s",
+ "▁na mes",
+ "▁nam es",
+ "▁ names",
+ "wh ere",
+ "whe re",
+ "w here",
+ "▁The se",
+ "▁Th ese",
+ "▁ These",
+ "▁s il",
+ "▁si l",
+ "▁ sil",
+ "é t",
+ "▁e ner",
+ "▁en er",
+ "▁ ener",
+ "▁N ow",
+ "▁No w",
+ "▁ Now",
+ "▁add ress",
+ "▁addr ess",
+ "▁ address",
+ "Res ponse",
+ "▁M r",
+ "▁ Mr",
+ "▁an sw",
+ "▁ans w",
+ "▁fil m",
+ "▁fi lm",
+ "▁ film",
+ "▁str ong",
+ "▁stro ng",
+ "▁ strong",
+ "▁b ring",
+ "▁br ing",
+ "▁Un ited",
+ "▁Unit ed",
+ "▁g e",
+ "▁ ge",
+ "▁w oman",
+ "▁wom an",
+ "▁wo man",
+ "▁ woman",
+ "Ne w",
+ "N ew",
+ "et t",
+ "e tt",
+ ". )",
+ "en ame",
+ "ena me",
+ "e name",
+ "▁A N",
+ "▁ AN",
+ "▁de scrib",
+ "▁desc rib",
+ "з а",
+ "is ing",
+ "isi ng",
+ "i sing",
+ "E L",
+ "q l",
+ "▁f ur",
+ "▁fu r",
+ "▁ fur",
+ "y ing",
+ "▁C al",
+ "▁Ca l",
+ "▁ Cal",
+ "▁D r",
+ "▁ Dr",
+ "ER R",
+ "E RR",
+ "▁\\ \\",
+ "▁ \\\\",
+ "an gle",
+ "ang le",
+ "ur ope",
+ "uro pe",
+ "urop e",
+ "▁c ity",
+ "▁cit y",
+ "▁ci ty",
+ "▁ city",
+ "▁in dex",
+ "▁ind ex",
+ "▁inde x",
+ "▁ index",
+ "▁a ction",
+ "▁act ion",
+ "▁ action",
+ "▁How ever",
+ "▁ However",
+ "▁f ig",
+ "▁fi g",
+ "▁ fig",
+ "ia s",
+ "i as",
+ "▁quest ion",
+ "▁ question",
+ "▁J an",
+ "▁Ja n",
+ "▁ Jan",
+ "▁M ed",
+ "▁Me d",
+ "▁ Med",
+ "▁C ont",
+ "▁Con t",
+ "▁Co nt",
+ "▁ Cont",
+ "am ed",
+ "ame d",
+ "a med",
+ "Cal l",
+ "C all",
+ "pl ied",
+ "tt y",
+ "t ty",
+ "▁ind ivid",
+ "pa ge",
+ "pag e",
+ "p age",
+ "▁c omb",
+ "▁com b",
+ "▁co mb",
+ "▁ comb",
+ "se ction",
+ "sect ion",
+ "s ection",
+ "▁C omm",
+ "▁Com m",
+ "▁Co mm",
+ "▁ Comm",
+ "ue l",
+ "u el",
+ "▁h et",
+ "▁he t",
+ "▁ het",
+ "▁B ar",
+ "▁Ba r",
+ "▁ Bar",
+ "ag ement",
+ "age ment",
+ "agem ent",
+ "fi n",
+ "f in",
+ "▁m ajor",
+ "▁ma jor",
+ "▁maj or",
+ "▁ major",
+ "op er",
+ "ope r",
+ "o per",
+ "ap i",
+ "a pi",
+ "ro om",
+ "r oom",
+ "▁ „",
+ "▁h ab",
+ "▁ha b",
+ "▁ hab",
+ "з и",
+ "▁a uf",
+ "▁au f",
+ "▁ auf",
+ "cur rent",
+ "curr ent",
+ "n i",
+ "▁in clude",
+ "▁incl ude",
+ "▁includ e",
+ "▁inclu de",
+ "▁ include",
+ "▁qu i",
+ "▁q ui",
+ "v a",
+ "U E",
+ "▁ide a",
+ "▁id ea",
+ "▁ idea",
+ ", '",
+ "▁requ ired",
+ "▁require d",
+ "▁ required",
+ "▁he art",
+ "▁hear t",
+ "▁ heart",
+ "ib ility",
+ "ibil ity",
+ "ict ion",
+ "i ction",
+ "Mod el",
+ "Mode l",
+ "Mo del",
+ "wr ite",
+ "writ e",
+ "w rite",
+ "▁cont ent",
+ "▁conten t",
+ "▁ content",
+ "▁w er",
+ "▁we r",
+ "▁ wer",
+ "▁h ands",
+ "▁hand s",
+ "▁han ds",
+ "ze n",
+ "z en",
+ "ch ar",
+ "cha r",
+ "c har",
+ "}^ {",
+ "} ^{",
+ "▁m ass",
+ "▁ma ss",
+ "▁mas s",
+ "▁ mass",
+ "pl y",
+ "p ly",
+ "▁n at",
+ "▁na t",
+ "▁ nat",
+ "re l",
+ "r el",
+ "▁d at",
+ "▁da t",
+ "▁ dat",
+ "==== ============",
+ "======== ========",
+ "============ ====",
+ "im al",
+ "ima l",
+ "i mal",
+ "▁pro bably",
+ "▁prob ably",
+ "un ch",
+ "unc h",
+ "▁m er",
+ "▁me r",
+ "▁ mer",
+ "il ar",
+ "ila r",
+ "i lar",
+ "ir es",
+ "ire s",
+ "i res",
+ "▁w atch",
+ "▁wat ch",
+ "▁ watch",
+ "S I",
+ "▁c ult",
+ "▁cu lt",
+ "▁cul t",
+ "▁m other",
+ "▁mot her",
+ "▁mo ther",
+ "▁ mother",
+ "▁govern ment",
+ "or ding",
+ "ord ing",
+ "▁( )",
+ "▁ ()",
+ "▁p ri",
+ "▁pr i",
+ "▁l ink",
+ "▁lin k",
+ "▁ link",
+ "gr oup",
+ "gro up",
+ "g roup",
+ "O L",
+ "▁n ear",
+ "▁ne ar",
+ "▁S er",
+ "▁Se r",
+ "▁ Ser",
+ "Se r",
+ "S er",
+ "it o",
+ "i to",
+ "▁value s",
+ "▁val ues",
+ "▁ values",
+ "▁j ava",
+ "▁ja va",
+ "▁ java",
+ "ful ly",
+ "full y",
+ "f ully",
+ "Co unt",
+ "C ount",
+ "++ )",
+ "▁v i",
+ "▁ vi",
+ "▁wh ite",
+ "▁ white",
+ "ma t",
+ "m at",
+ "ct x",
+ "c tx",
+ "▁con c",
+ "▁co nc",
+ "▁ conc",
+ "▁st ay",
+ "▁sta y",
+ "gi ng",
+ "gin g",
+ "g ing",
+ "▁c lear",
+ "▁cl ear",
+ "▁cle ar",
+ "▁ clear",
+ "▁c opy",
+ "▁co py",
+ "▁cop y",
+ "▁ copy",
+ "sel ves",
+ "▁prov ide",
+ "▁w ords",
+ "▁wor ds",
+ "▁word s",
+ "▁ words",
+ "com p",
+ "co mp",
+ "c omp",
+ "ar gs",
+ "arg s",
+ "▁p ick",
+ "▁pi ck",
+ "▁pic k",
+ "▁ pick",
+ "ul y",
+ "u ly",
+ "▁v ari",
+ "▁var i",
+ "▁va ri",
+ "▁ vari",
+ "▁bel ieve",
+ "▁belie ve",
+ "▁C o",
+ "▁ Co",
+ "Pro perty",
+ "Gr oup",
+ "G roup",
+ "▁t en",
+ "▁te n",
+ "▁ ten",
+ "is chen",
+ "isch en",
+ "ische n",
+ "isc hen",
+ "i schen",
+ "et urn",
+ "e turn",
+ "iv al",
+ "iva l",
+ "i val",
+ "Sys tem",
+ "S ystem",
+ "C L",
+ "be d",
+ "b ed",
+ "▁t otal",
+ "▁to tal",
+ "▁tot al",
+ "▁ total",
+ "▁is t",
+ "▁i st",
+ "▁ ist",
+ "In put",
+ "um ents",
+ "ument s",
+ "umen ts",
+ "u ments",
+ "Man ager",
+ "ш и",
+ "▁w in",
+ "▁ win",
+ "le ep",
+ "lee p",
+ "P I",
+ "но го",
+ "н ого",
+ "ru ction",
+ "ruct ion",
+ "r uction",
+ "▁in te",
+ "▁i nte",
+ "▁int e",
+ "▁ inte",
+ "Ap p",
+ "A pp",
+ "av or",
+ "avo r",
+ "a vor",
+ "▁re spect",
+ "▁res pect",
+ "▁resp ect",
+ "▁ respect",
+ "at ors",
+ "ator s",
+ "ato rs",
+ "▁c omo",
+ "▁com o",
+ "▁co mo",
+ "▁c ut",
+ "▁cu t",
+ "▁ cut",
+ "F A",
+ "▁s us",
+ "▁su s",
+ "▁A pp",
+ "▁Ap p",
+ "▁ App",
+ "re ct",
+ "rec t",
+ "r ect",
+ "F I",
+ "▁be gan",
+ "▁beg an",
+ "op h",
+ "o ph",
+ "▁s ort",
+ "▁so rt",
+ "▁sor t",
+ "▁ sort",
+ "th ough",
+ "ј е",
+ "ic ro",
+ "i cro",
+ "Tr ans",
+ "Tra ns",
+ "л і",
+ "▁In st",
+ "▁Ins t",
+ "▁ Inst",
+ "re quest",
+ "requ est",
+ "req uest",
+ "о р",
+ "▁rel ations",
+ "▁relation s",
+ "- \\",
+ "St atus",
+ "Stat us",
+ "ж и",
+ "▁f ather",
+ "▁fa ther",
+ "▁fat her",
+ "▁ father",
+ "c s",
+ "▁s ex",
+ "▁se x",
+ "▁ sex",
+ "is ch",
+ "isc h",
+ "i sch",
+ "v o",
+ "}_ {",
+ "} _{",
+ "ave n",
+ "av en",
+ "a ven",
+ "▁N e",
+ "▁ Ne",
+ "AT E",
+ "A TE",
+ "it ten",
+ "itt en",
+ "itte n",
+ "▁e ss",
+ "▁es s",
+ "▁ ess",
+ "T H",
+ "ight s",
+ "igh ts",
+ "▁h om",
+ "▁ho m",
+ "▁ hom",
+ "▁t oday",
+ "▁to day",
+ "▁tod ay",
+ "▁toda y",
+ "▁z u",
+ "▁ zu",
+ "it a",
+ "i ta",
+ "▁is n",
+ "▁i sn",
+ "▁o pt",
+ "▁op t",
+ "▁ opt",
+ "og n",
+ "o gn",
+ "é r",
+ "▁wh ether",
+ "▁whe ther",
+ "ix ed",
+ "ph i",
+ "p hi",
+ "id ence",
+ "iden ce",
+ "al d",
+ "a ld",
+ "Cl ient",
+ "A t",
+ "▁de ath",
+ "▁L et",
+ "▁Le t",
+ "▁ Let",
+ "iu s",
+ "i us",
+ "г и",
+ "▁р е",
+ "▁ ре",
+ "be n",
+ "b en",
+ ") \r",
+ "b a",
+ ">< /",
+ "> ",
+ "ave l",
+ "av el",
+ "a vel",
+ "▁m iss",
+ "▁mis s",
+ "▁mi ss",
+ "▁ miss",
+ "▁n ode",
+ "▁no de",
+ "▁nod e",
+ "▁ node",
+ "▁( $",
+ "▁ ($",
+ "▁col or",
+ "▁co lor",
+ "▁ color",
+ "▁o bt",
+ "▁ob t",
+ "to t",
+ "t ot",
+ "▁п ре",
+ "▁пр е",
+ "▁ пре",
+ "CO N",
+ "C ON",
+ "et te",
+ "ett e",
+ "▁G o",
+ "▁ Go",
+ "F l",
+ "▁D on",
+ "▁Do n",
+ "▁ Don",
+ "▁c rit",
+ "▁cr it",
+ "▁cri t",
+ "▁ crit",
+ "▁r i",
+ "▁ ri",
+ "pos t",
+ "po st",
+ "p ost",
+ "▁- >",
+ "▁ ->",
+ "▁J ust",
+ "▁Ju st",
+ "▁ Just",
+ "Wh at",
+ "W hat",
+ "at al",
+ "ata l",
+ "a tal",
+ "▁M in",
+ "▁Mi n",
+ "▁ Min",
+ "▁C or",
+ "▁Co r",
+ "▁ Cor",
+ "▁d ark",
+ "▁dar k",
+ "▁ dark",
+ "r l",
+ "▁l arg",
+ "▁la rg",
+ "▁ larg",
+ "di ng",
+ "d ing",
+ "ó n",
+ "ou ch",
+ "o uch",
+ "▁u m",
+ "▁ um",
+ "▁e lect",
+ "▁el ect",
+ "▁ele ct",
+ "▁ elect",
+ "▁d am",
+ "▁da m",
+ "▁ dam",
+ "▁ne eds",
+ "▁need s",
+ "▁m atter",
+ "▁mat ter",
+ "▁matt er",
+ "▁r ather",
+ "▁rat her",
+ "▁ra ther",
+ "fr om",
+ "f rom",
+ "ra m",
+ "r am",
+ "▁ і",
+ "▁t aken",
+ "▁take n",
+ "▁tak en",
+ "▁ta ken",
+ "▁de al",
+ "▁per iod",
+ "▁ period",
+ "▁M on",
+ "▁Mo n",
+ "▁ Mon",
+ "▁ Л",
+ "▁A ug",
+ "▁Au g",
+ "▁ Aug",
+ "ru n",
+ "r un",
+ "m m",
+ "el le",
+ "ell e",
+ "e lle",
+ "▁ex port",
+ "▁exp ort",
+ "▁ export",
+ "S c",
+ "vi s",
+ "v is",
+ "ab or",
+ "a bor",
+ "▁aut hor",
+ "▁auth or",
+ "▁ author",
+ "è re",
+ "▁re member",
+ "▁rem ember",
+ "▁remem ber",
+ "▁re du",
+ "▁r edu",
+ "▁red u",
+ "▁ redu",
+ "▁L ist",
+ "▁Li st",
+ "▁Lis t",
+ "▁ List",
+ "▁f ocus",
+ "▁ focus",
+ "▁char acter",
+ "▁ character",
+ "Tab le",
+ "T able",
+ "▁individ ual",
+ "▁need ed",
+ "bu m",
+ "b um",
+ "▁st yle",
+ "▁sty le",
+ "▁ style",
+ "in ary",
+ "ina ry",
+ "inar y",
+ "ers ion",
+ "ou te",
+ "out e",
+ "o ute",
+ "▁P e",
+ "▁ Pe",
+ "▁h on",
+ "▁ho n",
+ "▁ hon",
+ "mu t",
+ "m ut",
+ "se e",
+ "s ee",
+ "▁bec ame",
+ "▁d ire",
+ "▁di re",
+ "▁dir e",
+ "▁ dire",
+ "▁d ocument",
+ "▁doc ument",
+ "▁ document",
+ "se c",
+ "s ec",
+ "en ing",
+ "eni ng",
+ "e ning",
+ "▁vis it",
+ "▁ visit",
+ "▁f ac",
+ "▁fa c",
+ "▁ fac",
+ "t x",
+ "do wn",
+ "d own",
+ "pl it",
+ "p lit",
+ "▁ph ys",
+ "▁ phys",
+ "it ting",
+ "itt ing",
+ "jo y",
+ "j oy",
+ "▁h ig",
+ "▁hi g",
+ "Th is",
+ "T his",
+ "A d",
+ "▁B rit",
+ "▁Br it",
+ "▁em ploy",
+ "▁r é",
+ "▁ ré",
+ "▁ т",
+ "l ambda",
+ "▁im pro",
+ "▁imp ro",
+ "▁B o",
+ "▁ Bo",
+ "id ing",
+ "idi ng",
+ "i ding",
+ "▁on line",
+ "▁ online",
+ "me m",
+ "m em",
+ "at form",
+ "▁W ar",
+ "▁Wa r",
+ "▁ War",
+ "▁c as",
+ "▁ca s",
+ "▁ cas",
+ "as ure",
+ "a sure",
+ "▁p ur",
+ "▁pu r",
+ "▁ pur",
+ "me di",
+ "med i",
+ "m edi",
+ "Di s",
+ "D is",
+ "▁G erm",
+ "▁Ge rm",
+ "▁Ger m",
+ "p c",
+ "с а",
+ "▁friend s",
+ "▁M c",
+ "▁ Mc",
+ "D I",
+ "▁pl us",
+ "▁ plus",
+ "▁S et",
+ "▁Se t",
+ "▁ Set",
+ "idd le",
+ "it ut",
+ "itu t",
+ "▁de pend",
+ "▁dep end",
+ "▁ depend",
+ "re st",
+ "res t",
+ "r est",
+ "▁J e",
+ "▁ Je",
+ "▁h or",
+ "▁ho r",
+ "▁ hor",
+ "▁ent ire",
+ "Qu ery",
+ "Que ry",
+ "▁re fer",
+ "▁ref er",
+ "▁ refer",
+ "▁h ot",
+ "▁ho t",
+ "▁ hot",
+ "▁A ust",
+ "▁Aus t",
+ "▁Au st",
+ "▁com mon",
+ "▁comm on",
+ "▁ common",
+ "ц і",
+ "▁p ull",
+ "▁pu ll",
+ "▁pul l",
+ "▁ pull",
+ "▁A dd",
+ "▁Ad d",
+ "▁ Add",
+ "▁se ason",
+ "▁sea son",
+ "▁seas on",
+ "▁ season",
+ "▁in vol",
+ "▁inv ol",
+ "▁W orld",
+ "▁Wor ld",
+ "▁ World",
+ "cl ient",
+ "cli ent",
+ "no w",
+ "n ow",
+ "tr ue",
+ "ap pend",
+ "app end",
+ "appe nd",
+ "appen d",
+ "it ted",
+ "itt ed",
+ "itte d",
+ "em pt",
+ "emp t",
+ ") {",
+ "// /",
+ "/ //",
+ "▁p rop",
+ "▁pro p",
+ "▁pr op",
+ "▁ prop",
+ "im ate",
+ "ima te",
+ "imat e",
+ "i mate",
+ "S C",
+ "▁h ours",
+ "▁hour s",
+ "▁ho urs",
+ "▁h ope",
+ "▁hop e",
+ "▁ho pe",
+ "an dom",
+ "and om",
+ "ando m",
+ "і д",
+ "ist ic",
+ "isti c",
+ "▁pro perty",
+ "▁proper ty",
+ "▁ property",
+ "s g",
+ "> (",
+ "▁w rite",
+ "▁wr ite",
+ "▁writ e",
+ "▁ write",
+ "mar k",
+ "m ark",
+ "fin d",
+ "fi nd",
+ "f ind",
+ "▁person al",
+ "▁pers onal",
+ "▁persona l",
+ "▁ personal",
+ "] [",
+ "ro wn",
+ "row n",
+ "r own",
+ "P h",
+ "▁f oot",
+ "▁fo ot",
+ "▁foo t",
+ "▁ foot",
+ "▁re search",
+ "▁res earch",
+ "iron ment",
+ "▁n om",
+ "▁no m",
+ "▁ nom",
+ "▁in stance",
+ "▁inst ance",
+ "▁ instance",
+ "▁h eld",
+ "▁he ld",
+ "▁hel d",
+ "▁ held",
+ "D e",
+ "▁mem bers",
+ "▁member s",
+ "▁ members",
+ "▁f ire",
+ "▁fi re",
+ "▁fir e",
+ "▁ fire",
+ "▁hist ory",
+ "▁histor y",
+ "▁hi story",
+ "▁ history",
+ "▁m ap",
+ "▁ma p",
+ "▁ map",
+ "▁dis cuss",
+ "▁disc uss",
+ "▁e spec",
+ "▁es pec",
+ "▁esp ec",
+ "▁ espec",
+ "▁t aking",
+ "▁tak ing",
+ "▁ta king",
+ "▁s ervices",
+ "▁serv ices",
+ "▁service s",
+ "▁ services",
+ "▁ind ust",
+ "▁indu st",
+ "▁ indust",
+ "ig en",
+ "ige n",
+ "i gen",
+ "▁A ss",
+ "▁As s",
+ "▁ Ass",
+ "▁e xpected",
+ "▁ex pected",
+ "▁expect ed",
+ "▁ expected",
+ "▁wur de",
+ "di r",
+ "d ir",
+ "▁a mong",
+ "▁am ong",
+ "▁s ugg",
+ "▁su gg",
+ "▁sug g",
+ "re c",
+ "r ec",
+ "In ter",
+ "Int er",
+ "bl ock",
+ "blo ck",
+ "b lock",
+ "▁R ep",
+ "▁Re p",
+ "▁ Rep",
+ "▁p ain",
+ "▁pa in",
+ "▁f ive",
+ "▁fi ve",
+ "▁ five",
+ "▁f und",
+ "▁fun d",
+ "▁fu nd",
+ "▁ fund",
+ "ri d",
+ "r id",
+ "ar row",
+ "arr ow",
+ "▁t reat",
+ "▁tre at",
+ "▁he ard",
+ "▁hear d",
+ "▁de term",
+ "▁det erm",
+ "▁deter m",
+ "ic ult",
+ "▁s ense",
+ "▁sens e",
+ "▁sen se",
+ "es e",
+ "e se",
+ "F un",
+ "▁month s",
+ "▁mont hs",
+ "js on",
+ "j son",
+ ", ”",
+ "T I",
+ "or age",
+ "ora ge",
+ "o rage",
+ "▁ У",
+ "▁every one",
+ "▁c los",
+ "▁cl os",
+ "▁clo s",
+ "▁ clos",
+ "ie rs",
+ "ier s",
+ "i ers",
+ "air s",
+ "ai rs",
+ "a irs",
+ "def ine",
+ "I f",
+ "os p",
+ "o sp",
+ "▁w onder",
+ "▁won der",
+ "▁wo nder",
+ "N A",
+ "qu ery",
+ "que ry",
+ "quer y",
+ "p g",
+ "it es",
+ "ite s",
+ "i tes",
+ "▁m aterial",
+ "▁mat erial",
+ "▁mate rial",
+ "▁mater ial",
+ "▁ material",
+ "y d",
+ "Re ad",
+ "R ead",
+ "ht ml",
+ "h tml",
+ "T E",
+ "P r",
+ "^{ \\",
+ "^ {\\",
+ "▁g ave",
+ "▁ga ve",
+ "▁I S",
+ "▁ IS",
+ "▁s uggest",
+ "▁sugg est",
+ "▁sug gest",
+ "Over ride",
+ "ro du",
+ "rod u",
+ "Fr om",
+ "F rom",
+ "▁E urope",
+ "▁Europ e",
+ "▁Euro pe",
+ "▁ Europe",
+ "P O",
+ "▁s oon",
+ "▁so on",
+ "ho st",
+ "hos t",
+ "h ost",
+ "▁B er",
+ "▁Be r",
+ "▁ Ber",
+ ".. ..",
+ "... .",
+ ". ...",
+ "▁H ar",
+ "▁Ha r",
+ "▁ Har",
+ "▁e nergy",
+ "▁ener gy",
+ "▁energ y",
+ "▁ energy",
+ "> <",
+ "ave s",
+ "av es",
+ "a ves",
+ "▁e asy",
+ "▁eas y",
+ "▁b re",
+ "▁br e",
+ "▁ bre",
+ "fr ame",
+ "▁g round",
+ "▁gr ound",
+ "▁gro und",
+ "▁ ground",
+ "wi th",
+ "w ith",
+ "▁in side",
+ "▁ins ide",
+ "ie f",
+ "i ef",
+ "▁m o",
+ "▁ mo",
+ "p m",
+ "pa n",
+ "p an",
+ "ig r",
+ "i gr",
+ "▁o m",
+ "▁ om",
+ "ne xt",
+ "nex t",
+ "n ext",
+ "om et",
+ "ome t",
+ "o met",
+ "▁st atus",
+ "▁stat us",
+ "▁ status",
+ "▁} \r",
+ "▁ }\r",
+ "▁mus ic",
+ "or a",
+ "o ra",
+ "il es",
+ "ile s",
+ "i les",
+ "k i",
+ "▁e sc",
+ "▁es c",
+ "▁ esc",
+ "▁b es",
+ "▁be s",
+ "▁ bes",
+ "▁D is",
+ "▁Di s",
+ "▁ Dis",
+ "▁h ost",
+ "▁ho st",
+ "▁ host",
+ "▁c omes",
+ "▁com es",
+ "▁co mes",
+ "▁come s",
+ "▁ comes",
+ "us ed",
+ "use d",
+ "u sed",
+ "▁f uture",
+ "▁fut ure",
+ "▁ future",
+ "lic k",
+ "li ck",
+ "l ick",
+ "ai d",
+ "a id",
+ "▁com pet",
+ "▁comp et",
+ "▁ compet",
+ "▁v oice",
+ "▁vo ice",
+ "▁ voice",
+ "▁l oad",
+ "▁lo ad",
+ "▁ load",
+ "ev el",
+ "eve l",
+ "e vel",
+ "▁n eg",
+ "▁ne g",
+ "▁ neg",
+ "▁com mand",
+ "▁comm and",
+ "▁ command",
+ "▁f ür",
+ "▁p ie",
+ "▁pi e",
+ "▁ pie",
+ "▁qu ite",
+ "▁qui te",
+ "▁quit e",
+ "▁b lo",
+ "▁bl o",
+ "▁ blo",
+ "ag n",
+ "a gn",
+ "il on",
+ "ilo n",
+ "i lon",
+ "▁cl aim",
+ "▁ claim",
+ "▁t each",
+ "▁te ach",
+ "▁tea ch",
+ "▁pre vious",
+ "▁prev ious",
+ "▁ previous",
+ "▁s ite",
+ "▁sit e",
+ "▁si te",
+ "▁ site",
+ "co lor",
+ "col or",
+ "colo r",
+ "at tr",
+ "att r",
+ "▁ac cept",
+ "▁ accept",
+ "▁ex act",
+ ") }",
+ "af t",
+ "a ft",
+ "rol ler",
+ "roll er",
+ "о н",
+ "o o",
+ "Dat e",
+ "Da te",
+ "D ate",
+ "▁o u",
+ "▁ ou",
+ "s y",
+ "▁pre tty",
+ "▁pret ty",
+ "▁im age",
+ "▁imag e",
+ "▁ image",
+ "B U",
+ "▁term s",
+ "▁ter ms",
+ "▁s earch",
+ "▁se arch",
+ "▁sear ch",
+ "▁ search",
+ "▁ è",
+ "▁V al",
+ "▁Va l",
+ "▁ Val",
+ "▁ ‘",
+ "▁D av",
+ "▁Da v",
+ "M S",
+ "sr c",
+ "s rc",
+ "ma r",
+ "m ar",
+ "in cip",
+ "inc ip",
+ "▁could n",
+ "ad os",
+ "ado s",
+ "▁d ro",
+ "▁dr o",
+ "▁ dro",
+ "be ta",
+ "bet a",
+ "b eta",
+ "im um",
+ "▁min utes",
+ "▁minute s",
+ "▁minut es",
+ "▁g rand",
+ "▁gr and",
+ "▁gran d",
+ "▁gra nd",
+ "▁ grand",
+ "▁ »",
+ "▁O ur",
+ "▁ Our",
+ "St r",
+ "S tr",
+ "VE R",
+ "V ER",
+ "ma z",
+ "m az",
+ "▁or iginal",
+ "▁orig inal",
+ "▁origin al",
+ "▁ original",
+ "in i",
+ "i ni",
+ "▁c oll",
+ "▁col l",
+ "▁co ll",
+ "▁ coll",
+ "lo at",
+ "▁o s",
+ "▁ os",
+ "}) ;",
+ "} );",
+ "sum mary",
+ "▁w all",
+ "▁wa ll",
+ "▁wal l",
+ "▁ wall",
+ "Col or",
+ "Co lor",
+ "▁v ers",
+ "▁ver s",
+ "▁ve rs",
+ "▁ vers",
+ "▁d ella",
+ "▁de lla",
+ "▁del la",
+ "▁dell a",
+ "▁\" \"\"",
+ "▁\"\" \"",
+ "▁ \"\"\"",
+ "math bf",
+ "ze r",
+ "z er",
+ "au r",
+ "a ur",
+ "▁tr ack",
+ "▁tra ck",
+ "▁ track",
+ "▁ass oci",
+ "▁ associ",
+ "▁s uff",
+ "▁su ff",
+ "▁in de",
+ "▁i nde",
+ "▁ind e",
+ "▁ inde",
+ "ag ue",
+ "agu e",
+ "a gue",
+ "▁A pr",
+ "▁Ap r",
+ "▁ Apr",
+ "L e",
+ "ro ups",
+ "rou ps",
+ "roup s",
+ "bo ard",
+ "b oard",
+ "▁att ack",
+ "▁s eries",
+ "▁se ries",
+ "▁ser ies",
+ "▁serie s",
+ "▁ series",
+ "▁in stead",
+ "▁inst ead",
+ "ha m",
+ "h am",
+ "bo ok",
+ "b ook",
+ "▁s ix",
+ "▁si x",
+ "▁ six",
+ "▁R ec",
+ "▁Re c",
+ "▁ Rec",
+ "▁c oming",
+ "▁com ing",
+ "▁co ming",
+ "▁ coming",
+ "ur t",
+ "u rt",
+ "▁gl obal",
+ "▁glob al",
+ "▁glo bal",
+ "▁ global",
+ "▁ne cess",
+ "▁neces s",
+ "▁ necess",
+ "le ge",
+ "leg e",
+ "Po s",
+ "P os",
+ "▁le ave",
+ "▁ leave",
+ "▁p od",
+ "▁po d",
+ "▁ pod",
+ "ateg ory",
+ "ategor y",
+ "u z",
+ "▁de ep",
+ "▁ deep",
+ "▁k m",
+ "▁ km",
+ "▁out side",
+ "▁outs ide",
+ "ha s",
+ "h as",
+ "opt ions",
+ "option s",
+ "o ptions",
+ "▁S m",
+ "▁ Sm",
+ "Su b",
+ "S ub",
+ "ro ws",
+ "row s",
+ "r ows",
+ "▁в и",
+ "▁ ви",
+ "▁St ates",
+ "▁State s",
+ "▁Stat es",
+ "▁Sta tes",
+ "▁ States",
+ "▁wr ong",
+ "▁how ever",
+ "▁s em",
+ "▁se m",
+ "▁ sem",
+ "▁c atch",
+ "▁cat ch",
+ "▁ catch",
+ "\") ,",
+ "\" ),",
+ "mod el",
+ "mode l",
+ "mo del",
+ "▁h ttp",
+ "▁htt p",
+ "▁ http",
+ "▁o ption",
+ "▁opt ion",
+ "▁ option",
+ "ri e",
+ "r ie",
+ "▁с та",
+ "▁ст а",
+ "▁ ста",
+ "▁ä r",
+ "▁ är",
+ "▁en joy",
+ "▁enjo y",
+ "n u",
+ "▁p as",
+ "▁pa s",
+ "▁ pas",
+ "▁a mount",
+ "▁am ount",
+ "▁ amount",
+ "▁res pons",
+ "▁respon s",
+ "▁resp ons",
+ "▁ respons",
+ "▁In tern",
+ "▁Inter n",
+ "▁Int ern",
+ "▁ Intern",
+ "▁my self",
+ "▁o pp",
+ "▁op p",
+ "▁ opp",
+ "▁S im",
+ "▁Si m",
+ "▁ Sim",
+ "▁s ens",
+ "▁se ns",
+ "▁sen s",
+ "E d",
+ "▁( \\",
+ "▁ (\\",
+ "▁stud ents",
+ "▁student s",
+ "но в",
+ "н ов",
+ "▁point s",
+ "▁ points",
+ "ar ning",
+ "arn ing",
+ "U P",
+ "el ling",
+ "ell ing",
+ "elli ng",
+ "▁c annot",
+ "▁can not",
+ "B e",
+ "▁l ength",
+ "▁le ngth",
+ "▁ length",
+ "nu ll",
+ "n ull",
+ "ui nt",
+ "u int",
+ "wi se",
+ "w ise",
+ "▁d ouble",
+ "▁dou ble",
+ "▁doub le",
+ "▁ double",
+ "ig e",
+ "i ge",
+ "is ta",
+ "ist a",
+ "i sta",
+ "▁est ab",
+ "▁es tab",
+ "▁esta b",
+ "an ch",
+ "anc h",
+ "▁a go",
+ "▁ag o",
+ "▁ ago",
+ "▁b ound",
+ "▁bo und",
+ "▁bou nd",
+ "▁ bound",
+ "▁f a",
+ "▁ fa",
+ "▁c lean",
+ "▁cle an",
+ "▁ clean",
+ "▁sim ple",
+ "▁simpl e",
+ "▁ simple",
+ "m i",
+ "#### ####",
+ "if ier",
+ "ifi er",
+ "▁Gener al",
+ "▁Gen eral",
+ "▁Gene ral",
+ "▁ General",
+ "▁se emed",
+ "▁see med",
+ "▁seem ed",
+ "en a",
+ "e na",
+ "▁a ge",
+ "▁ag e",
+ "▁ age",
+ "но й",
+ "end if",
+ "A A",
+ "▁c aus",
+ "▁ca us",
+ "▁e duc",
+ "▁ed uc",
+ "▁ educ",
+ "▁c ell",
+ "▁ce ll",
+ "▁cel l",
+ "▁ cell",
+ "Ge ner",
+ "Gen er",
+ "G ener",
+ "sp ace",
+ "s pace",
+ "▁Y our",
+ "▁You r",
+ "▁ Your",
+ "▁be aut",
+ "g t",
+ "▁l imit",
+ "▁li mit",
+ "▁lim it",
+ "▁ limit",
+ "▁d ate",
+ "▁da te",
+ "▁dat e",
+ "▁ date",
+ "Ut il",
+ "U til",
+ "▁N ational",
+ "▁Nat ional",
+ "▁Nation al",
+ "▁ National",
+ "ow s",
+ "o ws",
+ "pa t",
+ "p at",
+ "qu ad",
+ "▁o k",
+ "▁ ok",
+ "▁ И",
+ "ar th",
+ "art h",
+ "ha t",
+ "h at",
+ "▁comm unity",
+ "▁commun ity",
+ "ou l",
+ "o ul",
+ "▁e conom",
+ "▁ec onom",
+ "▁ econom",
+ "Com ponent",
+ "bo r",
+ "b or",
+ "us ion",
+ "▁be low",
+ "▁bel ow",
+ "ear ch",
+ "e arch",
+ "or es",
+ "ore s",
+ "o res",
+ "ba n",
+ "b an",
+ "▁Aug ust",
+ "▁fur ther",
+ "sig ma",
+ "s igma",
+ "▁h a",
+ "▁ ha",
+ "j i",
+ "▁com put",
+ "▁comp ut",
+ "▁ comput",
+ "г ра",
+ "▁N one",
+ "▁No ne",
+ "▁Non e",
+ "▁ None",
+ "▁t er",
+ "▁te r",
+ "▁ ter",
+ "▁any one",
+ "▁t ask",
+ "▁ta sk",
+ "▁ task",
+ "en te",
+ "ent e",
+ "e nte",
+ "pos ition",
+ "pp ed",
+ "ppe d",
+ "p ped",
+ "▁a us",
+ "▁au s",
+ "▁ aus",
+ "Att ribute",
+ "Attrib ute",
+ "re q",
+ "r eq",
+ "ad dr",
+ "add r",
+ "li ght",
+ "lig ht",
+ "l ight",
+ "ш е",
+ "▁a rm",
+ "▁ar m",
+ "▁ arm",
+ "co ver",
+ "cov er",
+ "c over",
+ "up port",
+ "upp ort",
+ "▁G l",
+ "▁ Gl",
+ "▁S an",
+ "▁Sa n",
+ "▁ San",
+ "▁wr iting",
+ "▁writ ing",
+ "▁ writing",
+ "▁l ost",
+ "▁lo st",
+ "▁los t",
+ "▁M ark",
+ "▁Mar k",
+ "▁ Mark",
+ "▁g re",
+ "▁gr e",
+ "▁ gre",
+ "TY PE",
+ "T YPE",
+ "▁S outh",
+ "▁So uth",
+ "▁Sou th",
+ "▁Sout h",
+ "▁ South",
+ "▁per fect",
+ "▁perf ect",
+ "▁pack age",
+ "▁ package",
+ "▁in fl",
+ "▁inf l",
+ "▁ infl",
+ "ha ps",
+ "h aps",
+ "▁A ng",
+ "▁An g",
+ "▁ Ang",
+ "res pon",
+ "resp on",
+ "ri s",
+ "r is",
+ "pt ember",
+ "pte mber",
+ "▁build ing",
+ "▁ building",
+ "VA L",
+ "V AL",
+ "fr ee",
+ "fre e",
+ "f ree",
+ "▁c e",
+ "▁ ce",
+ "H T",
+ "▁F rom",
+ "▁Fr om",
+ "▁Fro m",
+ "▁ From",
+ "d s",
+ "ro y",
+ "r oy",
+ "ach ine",
+ "achi ne",
+ "no wn",
+ "now n",
+ "n own",
+ "▁sa ying",
+ "▁say ing",
+ "▁б ы",
+ "▁ бы",
+ "o e",
+ "Re f",
+ "R ef",
+ "▁net work",
+ "▁ network",
+ "par ent",
+ "pa rent",
+ "pare nt",
+ "paren t",
+ "p arent",
+ "ug e",
+ "u ge",
+ "▁sim ilar",
+ "> \r",
+ "Build er",
+ "B uilder",
+ "▁l iving",
+ "▁li ving",
+ "▁liv ing",
+ "▁contin ue",
+ "▁continu e",
+ "▁ continue",
+ "an ger",
+ "ang er",
+ "ange r",
+ "▁R ed",
+ "▁Re d",
+ "▁ Red",
+ "▁h air",
+ "▁ha ir",
+ "an ced",
+ "ance d",
+ "anc ed",
+ "ia ns",
+ "ian s",
+ "i ans",
+ "▁d ead",
+ "▁de ad",
+ "▁ dead",
+ "▁bo olean",
+ "▁ boolean",
+ "ic ation",
+ "▁д е",
+ "▁ де",
+ "▁cl ient",
+ "▁ client",
+ "uc t",
+ "u ct",
+ "▁ •",
+ "S P",
+ "ol der",
+ "old er",
+ "п е",
+ "ud io",
+ "udi o",
+ "▁d eg",
+ "▁de g",
+ "▁ deg",
+ "as ing",
+ "asi ng",
+ "a sing",
+ "▁st ep",
+ "▁ste p",
+ "▁ step",
+ "▁p ers",
+ "▁per s",
+ "▁pe rs",
+ "▁ pers",
+ "ç ão",
+ "ob j",
+ "o z",
+ "ul a",
+ "u la",
+ "▁r ound",
+ "▁ro und",
+ "▁rou nd",
+ "▁ round",
+ "▁u pon",
+ "▁up on",
+ "▁re source",
+ "▁res ource",
+ "▁ resource",
+ "▁val id",
+ "▁ valid",
+ "▁I I",
+ "▁ II",
+ "bu g",
+ "b ug",
+ "st d",
+ "s td",
+ "▁a ng",
+ "▁an g",
+ "▁ ang",
+ "sp an",
+ "s pan",
+ "po l",
+ "p ol",
+ "ial og",
+ "ia log",
+ "▁p hot",
+ "▁ph ot",
+ "? '",
+ "D B",
+ "▁F in",
+ "▁Fi n",
+ "▁ Fin",
+ "V E",
+ "E m",
+ "▁c am",
+ "▁ca m",
+ "▁ cam",
+ "tar get",
+ "t arget",
+ "pe cted",
+ "pect ed",
+ "pec ted",
+ "He l",
+ "H el",
+ "▁u t",
+ "▁ ut",
+ "▁T est",
+ "▁Te st",
+ "▁Tes t",
+ "▁ Test",
+ "▁t own",
+ "▁to wn",
+ "▁tow n",
+ "▁ town",
+ "al ign",
+ "ali gn",
+ "▁we bs",
+ "▁web s",
+ "in ner",
+ "inn er",
+ "au gh",
+ "aug h",
+ "a ugh",
+ "▁ex cept",
+ "▁ except",
+ "▁init ial",
+ "▁initi al",
+ "▁ initial",
+ "en ty",
+ "ent y",
+ "lic h",
+ "li ch",
+ "l ich",
+ "▁A ut",
+ "▁Au t",
+ "▁ Aut",
+ "to p",
+ "t op",
+ "▁f ail",
+ "▁fa il",
+ "▁ fail",
+ "on a",
+ "o na",
+ "▁ben ef",
+ "an ks",
+ "ank s",
+ "is che",
+ "isch e",
+ "isc he",
+ "i sche",
+ ". *",
+ "▁sign ific",
+ "▁cont act",
+ "▁ contact",
+ "Re c",
+ "R ec",
+ "ar io",
+ "ari o",
+ "a rio",
+ "ot tom",
+ "ott om",
+ "otto m",
+ "▁rel ationship",
+ "▁relations hip",
+ "▁relation ship",
+ "]) ;",
+ "] );",
+ "▁Н а",
+ "▁ На",
+ "He ad",
+ "H ead",
+ "form at",
+ "for mat",
+ "▁é t",
+ "▁ ét",
+ "▁M ore",
+ "▁Mor e",
+ "▁Mo re",
+ "▁ More",
+ "act ory",
+ "actor y",
+ "port un",
+ "+ \\",
+ "▁sim ply",
+ "▁simpl y",
+ "▁e p",
+ "▁ ep",
+ "▁R uss",
+ "▁Ru ss",
+ "▁Rus s",
+ "n í",
+ "u a",
+ "er c",
+ "e rc",
+ "▁long er",
+ "▁lon ger",
+ "in ition",
+ "init ion",
+ "ect or",
+ "ec tor",
+ "e ctor",
+ "apt ion",
+ "a ption",
+ "▁prof ess",
+ "▁profes s",
+ "▁M us",
+ "▁Mu s",
+ "▁ Mus",
+ "il ities",
+ "ili ties",
+ "è s",
+ "▁A ct",
+ "▁Ac t",
+ "▁ Act",
+ "off set",
+ "offs et",
+ "▁i ll",
+ "▁il l",
+ "▁ ill",
+ "ba nd",
+ "ban d",
+ "b and",
+ "▁A g",
+ "▁ Ag",
+ "▁П о",
+ "▁ По",
+ "б и",
+ "cont ent",
+ "ic on",
+ "ico n",
+ "i con",
+ "▁work s",
+ "▁wor ks",
+ "▁ works",
+ "yn am",
+ "yna m",
+ "y nam",
+ "pl ement",
+ "ple ment",
+ "p lement",
+ "Res ource",
+ "Re source",
+ "Act ion",
+ "A ction",
+ "▁diff icult",
+ "▁W est",
+ "▁We st",
+ "▁Wes t",
+ "▁ West",
+ "▁v ideo",
+ "▁vide o",
+ "▁ video",
+ "▁T HE",
+ "▁TH E",
+ "▁ THE",
+ "▁de cl",
+ "▁dec l",
+ "▁ decl",
+ "on don",
+ "ond on",
+ "ondo n",
+ "de d",
+ "d ed",
+ "}{ \\",
+ "} {\\",
+ "oc r",
+ "o cr",
+ "▁C ity",
+ "▁Cit y",
+ "▁Ci ty",
+ "▁ City",
+ "▁ я",
+ "ue r",
+ "u er",
+ "c z",
+ "▁im ag",
+ "▁i mag",
+ "▁ imag",
+ "c r",
+ "et e",
+ "e te",
+ "id get",
+ "idge t",
+ "▁M od",
+ "▁Mo d",
+ "▁ Mod",
+ "▁for ward",
+ "▁ forward",
+ "▁p ict",
+ "▁pi ct",
+ "▁pic t",
+ "or ge",
+ "org e",
+ "▁sub ject",
+ "▁ subject",
+ "up date",
+ "at tle",
+ "att le",
+ "s a",
+ "▁A nt",
+ "▁An t",
+ "▁ Ant",
+ "▁r unning",
+ "▁run ning",
+ "▁ running",
+ "▁s al",
+ "▁sa l",
+ "▁ sal",
+ "con ne",
+ "conn e",
+ "c onne",
+ "▁out put",
+ "▁ output",
+ "ad ata",
+ "ada ta",
+ "a data",
+ "M L",
+ "Che ck",
+ "C heck",
+ "led ge",
+ "l edge",
+ "▁p aper",
+ "▁pa per",
+ "▁pap er",
+ "▁ paper",
+ "param s",
+ "par ams",
+ "para ms",
+ "av y",
+ "a vy",
+ "▁a f",
+ "▁ af",
+ "▁e ine",
+ "▁ein e",
+ "▁j our",
+ "▁jo ur",
+ "▁jou r",
+ "▁ jour",
+ "A Y",
+ "▁it self",
+ "▁its elf",
+ "▁S tr",
+ "▁St r",
+ "▁ Str",
+ "st yle",
+ "sty le",
+ "Th at",
+ "T hat",
+ "▁m illion",
+ "▁mill ion",
+ "▁l anguage",
+ "▁ language",
+ "O S",
+ "vi ng",
+ "vin g",
+ "v ing",
+ "▁м а",
+ "▁ ма",
+ "▁т о",
+ "▁ то",
+ ") (",
+ "▁b uy",
+ "▁bu y",
+ ". /",
+ "▁. ..",
+ "▁.. .",
+ "▁ ...",
+ "▁t ried",
+ "▁tr ied",
+ "▁tri ed",
+ "▁com pl",
+ "▁comp l",
+ "▁act iv",
+ "▁ activ",
+ "ap ped",
+ "app ed",
+ "appe d",
+ "a pped",
+ "But ton",
+ "B utton",
+ "To ken",
+ "Tok en",
+ "T oken",
+ "▁prov ided",
+ "▁provide d",
+ "ib er",
+ "ibe r",
+ "i ber",
+ "▁c reated",
+ "▁cre ated",
+ "▁create d",
+ "▁creat ed",
+ "▁ created",
+ "cur ity",
+ "c urity",
+ "En d",
+ "E nd",
+ "a ł",
+ "us ter",
+ "ust er",
+ "u ster",
+ "iz ing",
+ "izi ng",
+ "i zing",
+ "om b",
+ "o mb",
+ "▁s ich",
+ "▁si ch",
+ "▁com pon",
+ "▁comp on",
+ "▁S ee",
+ "▁Se e",
+ "▁ See",
+ "▁u int",
+ "▁ui nt",
+ "▁ uint",
+ "▁l abel",
+ "▁la bel",
+ "▁lab el",
+ "▁ label",
+ "vo l",
+ "v ol",
+ "ó w",
+ "oc ol",
+ "oco l",
+ "o col",
+ "▁re ceived",
+ "▁rece ived",
+ "▁receive d",
+ "▁in tern",
+ "▁int ern",
+ "▁inter n",
+ "▁inte rn",
+ "▁ intern",
+ "ц е",
+ "R un",
+ "▁r oad",
+ "▁ro ad",
+ "▁ road",
+ "▁O ct",
+ "▁ Oct",
+ "▁C omp",
+ "▁Com p",
+ "▁Co mp",
+ "▁ Comp",
+ "▁stud y",
+ "▁т е",
+ "▁ те",
+ "Ac t",
+ "A ct",
+ "▁t our",
+ "▁to ur",
+ "▁tou r",
+ "▁St ate",
+ "▁Stat e",
+ "▁Sta te",
+ "▁ State",
+ "▁ad ded",
+ "▁add ed",
+ "▁ added",
+ "htt ps",
+ "http s",
+ "st ream",
+ "stre am",
+ "▁l ower",
+ "▁lo wer",
+ "▁low er",
+ "▁ lower",
+ "▁b ox",
+ "▁bo x",
+ "▁ box",
+ "▁S k",
+ "▁ Sk",
+ "▁them selves",
+ "▁c ross",
+ "▁cr oss",
+ "▁cro ss",
+ "▁ cross",
+ "▁e cho",
+ "▁ec ho",
+ "▁ echo",
+ "▁dev ice",
+ "▁ device",
+ "pos e",
+ "po se",
+ "p ose",
+ "▁g ames",
+ "▁game s",
+ "▁gam es",
+ "▁ga mes",
+ "P L",
+ "W indow",
+ "is es",
+ "ise s",
+ "i ses",
+ "ti tle",
+ "tit le",
+ "t itle",
+ "St ream",
+ "z t",
+ "▁S w",
+ "▁ Sw",
+ "▁r ole",
+ "▁ro le",
+ "▁ role",
+ "ia nt",
+ "ian t",
+ "i ant",
+ "k u",
+ "se qu",
+ "seq u",
+ "s equ",
+ "▁l ate",
+ "▁la te",
+ "▁lat e",
+ "▁ late",
+ "▁s old",
+ "▁so ld",
+ "▁sol d",
+ "р я",
+ "Com m",
+ "Co mm",
+ "C omm",
+ "▁en tre",
+ "▁ent re",
+ "▁entr e",
+ "▁ entre",
+ "▁d og",
+ "▁do g",
+ "▁ dog",
+ "dev ice",
+ "P ar",
+ "▁like ly",
+ "▁lik ely",
+ "▁ likely",
+ "^{ -",
+ "^ {-",
+ "▁l en",
+ "▁le n",
+ "▁ len",
+ "▁P aul",
+ "▁Pa ul",
+ "▁ Paul",
+ "▁t ool",
+ "▁to ol",
+ "▁too l",
+ "▁ tool",
+ "Of f",
+ "O ff",
+ "▁f amil",
+ "▁fam il",
+ "▁fa mil",
+ "▁d raw",
+ "▁dr aw",
+ "▁ draw",
+ "ap ping",
+ "app ing",
+ "a pping",
+ "▁ev ents",
+ "▁even ts",
+ "▁event s",
+ "▁ events",
+ "cre t",
+ "cr et",
+ "c ret",
+ "rou ght",
+ "rough t",
+ "r ought",
+ "Cont ent",
+ "▁soft ware",
+ "ri a",
+ "r ia",
+ "ms g",
+ "m sg",
+ "ga mma",
+ "g amma",
+ "▁h ear",
+ "▁he ar",
+ "Op er",
+ "O per",
+ "▁your self",
+ "▁yours elf",
+ "▁l iter",
+ "▁li ter",
+ "▁lit er",
+ "▁ liter",
+ "em p",
+ "e mp",
+ "▁se par",
+ "▁sep ar",
+ "▁ separ",
+ "▁ З",
+ "▁t itle",
+ "▁tit le",
+ "▁ti tle",
+ "▁ title",
+ "M ethod",
+ "math rm",
+ "▁s low",
+ "▁sl ow",
+ "▁R om",
+ "▁Ro m",
+ "▁ Rom",
+ "! !",
+ "▁t ax",
+ "▁ta x",
+ "▁ tax",
+ "ск а",
+ "с ка",
+ "empl ate",
+ "emp late",
+ "o i",
+ "▁A rt",
+ "▁Ar t",
+ "▁ Art",
+ "f alse",
+ "ast ic",
+ "ст ь",
+ "с ть",
+ "oc ket",
+ "ock et",
+ "▁e ns",
+ "▁en s",
+ "▁ ens",
+ "T O",
+ "am ente",
+ "ame nte",
+ "ament e",
+ "amen te",
+ "a mente",
+ "lo cal",
+ "loc al",
+ "l ocal",
+ "ch ie",
+ "chi e",
+ "▁p an",
+ "▁pa n",
+ "▁ pan",
+ "ни й",
+ "ch ema",
+ "che ma",
+ "chem a",
+ "▁N orth",
+ "▁Nor th",
+ "▁Nort h",
+ "з о",
+ "▁> =",
+ "▁ >=",
+ "A ut",
+ "▁d ig",
+ "▁di g",
+ "▁ dig",
+ "▁se ems",
+ "▁see ms",
+ "▁seem s",
+ "▁mor ning",
+ "so le",
+ "sol e",
+ "s ole",
+ "um er",
+ "ume r",
+ "u mer",
+ "del ta",
+ "d elta",
+ "it é",
+ "i té",
+ "ab ase",
+ "aba se",
+ "a base",
+ "ra f",
+ "r af",
+ "▁ob serv",
+ "▁obs erv",
+ "▁ observ",
+ "▁E st",
+ "▁Es t",
+ "▁ Est",
+ "▁s eg",
+ "▁se g",
+ "▁ seg",
+ "▁[ ]",
+ "▁ []",
+ "▁P res",
+ "▁Pr es",
+ "▁Pre s",
+ "▁ Pres",
+ "if ul",
+ "i ful",
+ "pu sh",
+ "pus h",
+ "p ush",
+ "▁O ff",
+ "▁Of f",
+ "▁ Off",
+ "ip e",
+ "i pe",
+ "at i",
+ "a ti",
+ "▁d im",
+ "▁di m",
+ "▁ dim",
+ "ce ed",
+ "c eed",
+ "En t",
+ "E nt",
+ "__ __",
+ "___ _",
+ "_ ___",
+ "en try",
+ "ent ry",
+ "entr y",
+ "▁f ight",
+ "▁fig ht",
+ "▁fi ght",
+ "▁c red",
+ "▁cre d",
+ "▁cr ed",
+ "▁ cred",
+ "▁O R",
+ "▁ OR",
+ "▁D ep",
+ "▁De p",
+ "▁ Dep",
+ "$ {",
+ "ле н",
+ "л ен",
+ "Creat e",
+ "C reate",
+ "▁Apr il",
+ "▁Ap ril",
+ "min istr",
+ "F L",
+ "▁A p",
+ "▁ Ap",
+ "▁H ere",
+ "▁He re",
+ "▁Her e",
+ "▁ Here",
+ "priv ate",
+ "p rivate",
+ "In stance",
+ "Inst ance",
+ "ie m",
+ "i em",
+ "▁off ice",
+ "▁offic e",
+ "▁th ird",
+ "▁ third",
+ "▁up date",
+ "▁ update",
+ "Lin e",
+ "Li ne",
+ "L ine",
+ "ta g",
+ "t ag",
+ "▁e specially",
+ "▁espec ially",
+ "▁especial ly",
+ "▁ especially",
+ "▁го да",
+ "▁год а",
+ "▁c u",
+ "▁ cu",
+ "▁k ill",
+ "▁kil l",
+ "▁ki ll",
+ "▁ kill",
+ "au ght",
+ "augh t",
+ "aug ht",
+ "▁s we",
+ "▁sw e",
+ "Option s",
+ "Opt ions",
+ "O ptions",
+ "I M",
+ "C C",
+ "▁com pan",
+ "▁comp an",
+ "ju st",
+ "j ust",
+ "▁Wh ile",
+ "▁ While",
+ "iz er",
+ "ize r",
+ "i zer",
+ "▁м о",
+ "▁ мо",
+ "к е",
+ "▁a uto",
+ "▁aut o",
+ "▁au to",
+ "▁ auto",
+ "▁b and",
+ "▁ban d",
+ "▁ba nd",
+ "▁ band",
+ "ме н",
+ "м ен",
+ "ique s",
+ "iqu es",
+ "iq ues",
+ "i ques",
+ "▁p le",
+ "▁pl e",
+ "▁ ple",
+ "N O",
+ "▁O F",
+ "▁ OF",
+ "▁s ong",
+ "▁so ng",
+ "▁son g",
+ "▁A cc",
+ "▁Ac c",
+ "▁ Acc",
+ "EX T",
+ "E XT",
+ "en sor",
+ "ens or",
+ "enso r",
+ "in ing",
+ "ini ng",
+ "i ning",
+ "▁l at",
+ "▁la t",
+ "▁ lat",
+ "bi g",
+ "b ig",
+ "▁K ing",
+ "▁Ki ng",
+ "▁Kin g",
+ "▁ King",
+ "oc h",
+ "o ch",
+ "s i",
+ "▁H ist",
+ "▁His t",
+ "▁Hi st",
+ "▁ Hist",
+ "▁qu ality",
+ "▁qual ity",
+ "▁ quality",
+ "mod e",
+ "mo de",
+ "m ode",
+ "▁op portun",
+ "▁would n",
+ ":* *",
+ ": **",
+ "out put",
+ "▁fe et",
+ "▁fee t",
+ "▁m is",
+ "▁mi s",
+ "d f",
+ "ag ing",
+ "agi ng",
+ "a ging",
+ "▁м е",
+ "▁ ме",
+ "▁t ro",
+ "▁tr o",
+ "▁d efined",
+ "▁def ined",
+ "▁define d",
+ "▁defin ed",
+ "▁ defined",
+ "▁re view",
+ "▁rev iew",
+ "▁ review",
+ "▁F il",
+ "▁Fi l",
+ "▁ Fil",
+ "> >",
+ "▁pr incip",
+ "▁prin cip",
+ "Bas e",
+ "B ase",
+ "di ct",
+ "d ict",
+ "ve rage",
+ "ver age",
+ "ic ient",
+ "ici ent",
+ "I F",
+ "▁h it",
+ "▁hi t",
+ "▁ hit",
+ "Pag e",
+ "P age",
+ "▁p erm",
+ "▁per m",
+ "▁pe rm",
+ "▁ perm",
+ "ce l",
+ "c el",
+ "í t",
+ "▁ex press",
+ "▁exp ress",
+ "▁expr ess",
+ "▁ express",
+ "▁ind ic",
+ "▁Se ptember",
+ "▁Sept ember",
+ "im age",
+ "ima ge",
+ "imag e",
+ "▁product s",
+ "▁ products",
+ "▁m edia",
+ "▁med ia",
+ "▁medi a",
+ "▁ media",
+ "ch ange",
+ "chan ge",
+ "ig ger",
+ "igg er",
+ "▁s end",
+ "▁se nd",
+ "▁sen d",
+ "▁ send",
+ "la st",
+ "las t",
+ "l ast",
+ "min g",
+ "mi ng",
+ "m ing",
+ "p a",
+ "ua ry",
+ "uar y",
+ "u ary",
+ "▁spe ak",
+ "ны й",
+ "щ е",
+ "ys is",
+ "y sis",
+ "ly ing",
+ "l ying",
+ "▁ ч",
+ "li ke",
+ "lik e",
+ "l ike",
+ "р ы",
+ "в і",
+ "▁M ich",
+ "▁Mic h",
+ "▁Mi ch",
+ "M O",
+ "▁J ah",
+ "▁Ja h",
+ "ens ive",
+ "▁sh are",
+ "▁shar e",
+ "▁sha re",
+ "▁ share",
+ "▁develop ment",
+ "C P",
+ "sp ec",
+ "spe c",
+ "s pec",
+ "▁f ast",
+ "▁fa st",
+ "▁ fast",
+ "he t",
+ "h et",
+ "H O",
+ "▁part icip",
+ "▁partic ip",
+ "▁parti cip",
+ "Bl ock",
+ "Blo ck",
+ "B lock",
+ "▁vi ol",
+ "▁fr ame",
+ "▁fra me",
+ "▁fram e",
+ "▁ frame",
+ "▁qu al",
+ "▁q ual",
+ "▁ qual",
+ "tr e",
+ "t re",
+ "▁ Ф",
+ "▁to ward",
+ "▁tow ard",
+ "f g",
+ "Bo x",
+ "B ox",
+ "Col umn",
+ "▁mil it",
+ "▁mi lit",
+ "▁M arch",
+ "▁Mar ch",
+ "▁Marc h",
+ "▁var ious",
+ "▁vari ous",
+ "pa ss",
+ "pas s",
+ "p ass",
+ "▁P ark",
+ "▁Par k",
+ "▁B en",
+ "▁Be n",
+ "▁ Ben",
+ "Fr ame",
+ "▁n ormal",
+ "▁nor mal",
+ "▁norm al",
+ "▁ normal",
+ "op en",
+ "ope n",
+ "o pen",
+ "p x",
+ "▁ph one",
+ "▁ phone",
+ "▁E ven",
+ "▁Ev en",
+ "▁Eve n",
+ "▁ Even",
+ "▁m a",
+ "▁ ma",
+ "ibr ary",
+ "St art",
+ "Star t",
+ "id den",
+ "idd en",
+ "rh o",
+ "r ho",
+ "gr aph",
+ "gra ph",
+ "g raph",
+ "ac ing",
+ "aci ng",
+ "a cing",
+ "' .",
+ "ar ter",
+ "art er",
+ "arte r",
+ "me s",
+ "m es",
+ "in st",
+ "ins t",
+ "▁i r",
+ "▁ ir",
+ "act ive",
+ "activ e",
+ "▁f em",
+ "▁fe m",
+ "▁ fem",
+ "▁m oved",
+ "▁mov ed",
+ "▁move d",
+ "▁mo ved",
+ "▁st ore",
+ "▁stor e",
+ "▁sto re",
+ "▁ store",
+ "▁p rice",
+ "▁pr ice",
+ "▁pri ce",
+ "▁ price",
+ "\") .",
+ "\" ).",
+ "ber g",
+ "be rg",
+ "b erg",
+ "▁n ov",
+ "▁no v",
+ "▁ nov",
+ "▁c ard",
+ "▁car d",
+ "▁ca rd",
+ "▁ card",
+ "el low",
+ "ell ow",
+ "ello w",
+ "▁part y",
+ "▁par ty",
+ "▁ party",
+ "▁M or",
+ "▁Mo r",
+ "ae l",
+ "a el",
+ "▁per cent",
+ "▁ percent",
+ "▁tr aining",
+ "▁tra ining",
+ "▁train ing",
+ "▁ training",
+ "▁in g",
+ "▁i ng",
+ "▁ ing",
+ "im er",
+ "ime r",
+ "i mer",
+ "▁S am",
+ "▁Sa m",
+ "▁ Sam",
+ "Def ault",
+ "▁f uck",
+ "▁fu ck",
+ "▁com plete",
+ "▁comp lete",
+ "▁complet e",
+ "▁compl ete",
+ "▁ complete",
+ "ui d",
+ "u id",
+ "▁det ails",
+ "▁detail s",
+ "▁ details",
+ "▁l ed",
+ "▁le d",
+ "▁ led",
+ "Po int",
+ "P oint",
+ "▁C ount",
+ "▁Co unt",
+ "▁Coun t",
+ "▁Cou nt",
+ "▁ Count",
+ "▁reg ard",
+ "z o",
+ "▁B ro",
+ "▁Br o",
+ "▁ Bro",
+ "▁rec ogn",
+ "▁ recogn",
+ "▁H ol",
+ "▁Ho l",
+ "▁ Hol",
+ "U M",
+ "el ement",
+ "ele ment",
+ "elem ent",
+ "e lement",
+ "Mod e",
+ "Mo de",
+ "M ode",
+ "▁ex am",
+ "▁E X",
+ "▁ EX",
+ "Im age",
+ "ver se",
+ "vers e",
+ "ri ter",
+ "rit er",
+ "rite r",
+ "r iter",
+ "so ft",
+ "s oft",
+ "▁int rodu",
+ "▁intro du",
+ "▁sur pr",
+ "Buf fer",
+ "Buff er",
+ "B uffer",
+ "le ctor",
+ "lect or",
+ "l ector",
+ "ar en",
+ "are n",
+ "a ren",
+ "an ged",
+ "ang ed",
+ "ange d",
+ "▁P at",
+ "▁Pa t",
+ "▁ Pat",
+ "▁P al",
+ "▁Pa l",
+ "▁ Pal",
+ "▁con tr",
+ "▁cont r",
+ "▁ contr",
+ "Hand ler",
+ "Handle r",
+ "▁fe atures",
+ "▁feature s",
+ "▁feat ures",
+ "▁ features",
+ "ip le",
+ "i ple",
+ "▁C ON",
+ "▁CO N",
+ "▁ CON",
+ "Fi l",
+ "F il",
+ "▁P ort",
+ "▁Po rt",
+ "▁Por t",
+ "▁ Port",
+ "▁th inking",
+ "▁think ing",
+ "▁thin king",
+ "do c",
+ "d oc",
+ "we r",
+ "w er",
+ "▁work ed",
+ "▁wor ked",
+ "P C",
+ "c m",
+ "da t",
+ "d at",
+ "PR O",
+ "P RO",
+ "▁E very",
+ "▁Ev ery",
+ "▁Ever y",
+ "▁Eve ry",
+ "▁ Every",
+ "▁e ra",
+ "▁er a",
+ "▁ era",
+ "▁F irst",
+ "▁ First",
+ "g n",
+ "▁im medi",
+ "▁imm edi",
+ "ov ember",
+ "ove mber",
+ "ap an",
+ "apa n",
+ "a pan",
+ "▁ex tra",
+ "▁ext ra",
+ "▁extr a",
+ "▁ extra",
+ "▁s ection",
+ "▁se ction",
+ "▁sect ion",
+ "▁ section",
+ "▁J une",
+ "▁Jun e",
+ "▁Ju ne",
+ "▁v ia",
+ "▁vi a",
+ "▁ via",
+ "▁g one",
+ "▁go ne",
+ "com e",
+ "co me",
+ "c ome",
+ "▁s tri",
+ "▁st ri",
+ "▁str i",
+ "▁ stri",
+ "^ \\",
+ "ant ly",
+ "▁ar ch",
+ "▁arc h",
+ "▁ arch",
+ "S ource",
+ "▁con v",
+ "▁co nv",
+ "▁ conv",
+ "▁L ondon",
+ "▁Lond on",
+ "▁ London",
+ "Num ber",
+ "N umber",
+ "▁quest ions",
+ "▁question s",
+ "an did",
+ "and id",
+ "▁play ed",
+ "en v",
+ "e nv",
+ "▁Sch ool",
+ "▁nat ural",
+ "▁natur al",
+ "▁ natural",
+ "ca n",
+ "c an",
+ "▁ne ws",
+ "▁new s",
+ "▁ news",
+ "D R",
+ "▁c hall",
+ "▁ch all",
+ "▁cha ll",
+ "▁S oc",
+ "▁So c",
+ "▁ э",
+ "▁att empt",
+ "* }",
+ "N ull",
+ "ro te",
+ "rot e",
+ "r ote",
+ "▁b i",
+ "▁ bi",
+ "▁wr itten",
+ "▁writ ten",
+ "▁ written",
+ "▁bl ood",
+ "▁blo od",
+ "▁happ ened",
+ "▁happen ed",
+ "▁c ause",
+ "▁caus e",
+ "▁ca use",
+ "as hing",
+ "ash ing",
+ "ashi ng",
+ "▁Will iam",
+ "ad em",
+ "ade m",
+ "a dem",
+ "▁b rought",
+ "▁br ought",
+ "▁dis play",
+ "▁displ ay",
+ "▁disp lay",
+ "▁ display",
+ "im a",
+ "i ma",
+ "▁fin ally",
+ "▁final ly",
+ "ta b",
+ "t ab",
+ "▁return ed",
+ "ны х",
+ "ni e",
+ "n ie",
+ "▁ q",
+ "▁h ers",
+ "▁he rs",
+ "▁her s",
+ "▁P re",
+ "▁Pr e",
+ "▁ Pre",
+ "▁d ou",
+ "▁do u",
+ "buf fer",
+ "buff er",
+ "b uffer",
+ "▁eff ort",
+ "ain e",
+ "ai ne",
+ "a ine",
+ "x y",
+ "▁his tor",
+ "▁hist or",
+ "en u",
+ "e nu",
+ "▁ar riv",
+ "▁arr iv",
+ "▁D em",
+ "▁De m",
+ "▁ Dem",
+ "▁f avor",
+ "▁fa vor",
+ "▁fav or",
+ "▁hand le",
+ "▁ handle",
+ "SE T",
+ "S ET",
+ "▁P ublic",
+ "▁Pub lic",
+ "▁Pu blic",
+ "▁ Public",
+ "ru pt",
+ "rup t",
+ "r upt",
+ "▁u r",
+ "▁ ur",
+ "▁for ce",
+ "▁ force",
+ "▁é s",
+ "▁ és",
+ "ub e",
+ "u be",
+ "Pr e",
+ "P re",
+ "р і",
+ "in y",
+ "i ny",
+ "th eta",
+ "the ta",
+ "is f",
+ "i sf",
+ "▁n ational",
+ "▁nat ional",
+ "▁nation al",
+ "Equ al",
+ "Eq ual",
+ "E qual",
+ "ren ch",
+ "▁w ife",
+ "▁c apt",
+ "▁cap t",
+ "▁ca pt",
+ "▁In ter",
+ "▁Int er",
+ "▁ Inter",
+ "ta u",
+ "t au",
+ "▁s leep",
+ "▁sle ep",
+ "▁ sleep",
+ "../ ../",
+ "▁iss ue",
+ "▁ issue",
+ "▁m ember",
+ "▁me mber",
+ "▁mem ber",
+ "▁ member",
+ "▁a wait",
+ "▁aw ait",
+ "▁ await",
+ "▁D an",
+ "▁Da n",
+ "▁ Dan",
+ "z i",
+ "in ate",
+ "ina te",
+ "i nate",
+ "▁s ym",
+ "▁sy m",
+ "▁ sym",
+ "ch an",
+ "cha n",
+ "c han",
+ "▁J ack",
+ "▁Jac k",
+ "▁Ja ck",
+ "▁ Jack",
+ "▁Eng lish",
+ "▁ English",
+ "▁s z",
+ "▁ sz",
+ "rib utes",
+ "ribut es",
+ "ribute s",
+ "ribu tes",
+ "▁i gn",
+ "▁ig n",
+ "▁ ign",
+ "á l",
+ "▁app ear",
+ "▁appe ar",
+ "ra d",
+ "r ad",
+ "id ge",
+ "▁co uple",
+ "▁cou ple",
+ "▁coup le",
+ "▁s hip",
+ "▁sh ip",
+ "▁ ship",
+ "li g",
+ "l ig",
+ "we b",
+ "w eb",
+ "▁us ually",
+ "▁usual ly",
+ "▁re ady",
+ "▁read y",
+ "▁ ready",
+ "▁v ill",
+ "▁vi ll",
+ "▁vil l",
+ "▁W hy",
+ "▁Wh y",
+ "▁ Why",
+ "eb ru",
+ "e bru",
+ "▁g rad",
+ "▁gr ad",
+ "▁gra d",
+ "▁ grad",
+ "or ds",
+ "ord s",
+ "▁in f",
+ "▁i nf",
+ "▁ inf",
+ "▁l oss",
+ "▁lo ss",
+ "▁los s",
+ "▁ loss",
+ "▁o d",
+ "▁ od",
+ "▁Ph il",
+ "▁ Phil",
+ "ser ver",
+ "serv er",
+ "serve r",
+ "▁U p",
+ "▁ Up",
+ "▁b uff",
+ "▁bu ff",
+ "▁buf f",
+ "▁ buff",
+ "▁fil ename",
+ "▁file name",
+ "▁ filename",
+ "AB LE",
+ "it ing",
+ "iti ng",
+ "i ting",
+ "ef ore",
+ "e fore",
+ "() ->",
+ "( )->",
+ "▁cond itions",
+ "▁condition s",
+ "▁ conditions",
+ "v m",
+ "el d",
+ "e ld",
+ "it z",
+ "i tz",
+ "▁Tr ans",
+ "▁Tra ns",
+ "▁ Trans",
+ "▁w eight",
+ "▁we ight",
+ "▁weigh t",
+ "▁ weight",
+ "▁high er",
+ "▁hig her",
+ "▁r ate",
+ "▁rat e",
+ "▁ra te",
+ "▁ rate",
+ "▁acc om",
+ "▁ac com",
+ "vi der",
+ "vid er",
+ "v ider",
+ "O M",
+ "▁w ays",
+ "▁way s",
+ "▁wa ys",
+ "▁ ways",
+ "com ing",
+ "co ming",
+ "c oming",
+ "▁l ock",
+ "▁loc k",
+ "▁lo ck",
+ "▁ lock",
+ "▁e tc",
+ "▁et c",
+ "▁ etc",
+ "▁a vec",
+ "▁av ec",
+ "▁ave c",
+ "▁t akes",
+ "▁take s",
+ "▁tak es",
+ "▁ta kes",
+ "▁C har",
+ "▁Ch ar",
+ "▁Cha r",
+ "▁ Char",
+ "▁N ovember",
+ "▁Nov ember",
+ "m ethod",
+ "▁A ustral",
+ "▁Aust ral",
+ "▁ Austral",
+ "▁Amer ica",
+ "▁ America",
+ "lo ng",
+ "lon g",
+ "l ong",
+ "ce mber",
+ "c ember",
+ "▁polit ical",
+ "fl ow",
+ "f low",
+ "▁may be",
+ "▁ maybe",
+ "▁a mb",
+ "▁am b",
+ "▁ amb",
+ "La yout",
+ "L ayout",
+ "il ed",
+ "ile d",
+ "i led",
+ "om en",
+ "ome n",
+ "o men",
+ "ol a",
+ "o la",
+ "ic ip",
+ "ici p",
+ "i cip",
+ "part ial",
+ "Tr ue",
+ "▁f loor",
+ "▁fl oor",
+ "▁flo or",
+ "▁ floor",
+ "▁D ef",
+ "▁De f",
+ "▁ Def",
+ "▁conc ern",
+ "▁conce rn",
+ "▁concer n",
+ "y r",
+ "▁sh ows",
+ "▁show s",
+ "i h",
+ "▁an swer",
+ "▁answ er",
+ "▁ans wer",
+ "▁ answer",
+ "ac c",
+ "a cc",
+ "▁b all",
+ "▁bal l",
+ "▁ba ll",
+ "▁ ball",
+ "▁R ev",
+ "▁Re v",
+ "▁ Rev",
+ "▁s un",
+ "▁su n",
+ "▁ sun",
+ "▁quick ly",
+ "▁s omet",
+ "▁so met",
+ "▁some t",
+ "▁som et",
+ "ment e",
+ "me nte",
+ "men te",
+ "m ente",
+ "▁M al",
+ "▁Ma l",
+ "▁ Mal",
+ "und red",
+ "▁iss ues",
+ "▁issue s",
+ "▁ issues",
+ "ec ause",
+ "eca use",
+ "pe s",
+ "p es",
+ "▁p layer",
+ "▁pl ayer",
+ "▁play er",
+ "▁ player",
+ "▁par ents",
+ "▁parent s",
+ "▁ parents",
+ "▁pop ular",
+ "▁popula r",
+ "▁popul ar",
+ "▁m ode",
+ "▁mod e",
+ "▁mo de",
+ "▁ mode",
+ "▁m ention",
+ "▁ment ion",
+ "N E",
+ "Lo ad",
+ "L oad",
+ "▁reg ular",
+ "▁regul ar",
+ "▁ regular",
+ "ave d",
+ "av ed",
+ "a ved",
+ "? :",
+ "ye ar",
+ "y ear",
+ "fun c",
+ "fu nc",
+ "f unc",
+ "▁per formance",
+ "▁perform ance",
+ "▁J uly",
+ "▁Jul y",
+ "▁Ju ly",
+ "th ern",
+ "ther n",
+ "the rn",
+ "▁we bsite",
+ "▁webs ite",
+ "▁web site",
+ "fo rd",
+ "for d",
+ "f ord",
+ "P R",
+ "el a",
+ "e la",
+ "le vel",
+ "lev el",
+ "l evel",
+ "ui t",
+ "u it",
+ "fl ags",
+ "flag s",
+ "▁w orth",
+ "▁wor th",
+ "▁ worth",
+ "▁cor respon",
+ "▁Brit ish",
+ "si m",
+ "s im",
+ "▁al one",
+ "▁ alone",
+ "▁h ar",
+ "▁ha r",
+ "▁ har",
+ "▁o nes",
+ "▁on es",
+ "▁one s",
+ "▁ ones",
+ "ob ile",
+ "obi le",
+ "obil e",
+ "▁d ru",
+ "▁dr u",
+ "▁ dru",
+ "ch i",
+ "c hi",
+ "▁D avid",
+ "▁Dav id",
+ "▁Da vid",
+ "▁ David",
+ "▁proble ms",
+ "▁problem s",
+ "▁col umn",
+ "▁ column",
+ "() ;\r",
+ "(); \r",
+ "( );\r",
+ "Z E",
+ "▁re lig",
+ "▁rel ig",
+ "▁reli g",
+ "olog ical",
+ "▁reg ion",
+ "▁ region",
+ "ad y",
+ "a dy",
+ "I O",
+ "an der",
+ "and er",
+ "ande r",
+ "a nder",
+ "Ne t",
+ "N et",
+ "▁bu ilt",
+ "▁ built",
+ "▁inst all",
+ "▁ install",
+ "▁appro ach",
+ "C ur",
+ "▁f ine",
+ "▁fin e",
+ "▁fi ne",
+ "▁talk ing",
+ "▁tal king",
+ "▁ch anges",
+ "▁chang es",
+ "▁change s",
+ "▁ changes",
+ "St yle",
+ "▁M art",
+ "▁Mar t",
+ "▁Ma rt",
+ "▁ Mart",
+ "л ю",
+ "res ponse",
+ "respon se",
+ "respons e",
+ "te ger",
+ "{ \r",
+ "ir it",
+ "iri t",
+ "i rit",
+ "▁prote cted",
+ "▁protect ed",
+ "▁ protected",
+ "▁re le",
+ "▁r ele",
+ "▁rel e",
+ "er ship",
+ "ers hip",
+ "те ль",
+ "тел ь",
+ "un signed",
+ "uns igned",
+ "ial ize",
+ "▁htt ps",
+ "▁http s",
+ "▁ https",
+ "T ag",
+ "▁$ (",
+ "▁ $(",
+ "mo re",
+ "mor e",
+ "m ore",
+ "ype s",
+ "yp es",
+ "y pes",
+ "▁st ream",
+ "▁stre am",
+ "▁ stream",
+ "et ch",
+ "etc h",
+ "▁eng ine",
+ "▁ engine",
+ "K E",
+ "cm d",
+ "c md",
+ "sc ript",
+ "scri pt",
+ "scr ipt",
+ "s cript",
+ "tt p",
+ "t tp",
+ "▁a void",
+ "▁av oid",
+ "▁t err",
+ "▁te rr",
+ "▁ter r",
+ "▁r ock",
+ "▁ro ck",
+ "▁ rock",
+ "▁f ul",
+ "▁fu l",
+ "▁ ful",
+ "Up date",
+ "▁env ironment",
+ "▁environ ment",
+ "▁ environment",
+ "▁p rec",
+ "▁pre c",
+ "▁pr ec",
+ "▁ prec",
+ "▁с а",
+ "▁ са",
+ "▁c ases",
+ "▁case s",
+ "▁cas es",
+ "▁ca ses",
+ "▁ cases",
+ "▁off set",
+ "▁ offset",
+ "▁r ais",
+ "▁ra is",
+ "▁ rais",
+ "li b",
+ "l ib",
+ "ée s",
+ "é es",
+ "a a",
+ "y t",
+ "▁a rr",
+ "▁ar r",
+ "▁ arr",
+ "opy right",
+ "f irst",
+ "▁u til",
+ "▁ut il",
+ "▁ util",
+ "▁fe ature",
+ "▁feat ure",
+ "▁ feature",
+ "pos ed",
+ "po sed",
+ "pose d",
+ "p osed",
+ "ff ect",
+ "f fect",
+ "ж а",
+ "it ude",
+ "itu de",
+ "itud e",
+ "em ents",
+ "ement s",
+ "emen ts",
+ "e ments",
+ "as c",
+ "a sc",
+ "ad or",
+ "ado r",
+ "le ctions",
+ "lect ions",
+ "lection s",
+ "▁cl ub",
+ "▁ club",
+ "] {",
+ "▁* )",
+ "▁ *)",
+ "ст во",
+ "ств о",
+ "с тво",
+ "▁im m",
+ "▁i mm",
+ "▁ imm",
+ "▁for mer",
+ "▁form er",
+ "▁forme r",
+ "▁ former",
+ "▁r ights",
+ "▁right s",
+ "▁dec ided",
+ "▁decide d",
+ "▁decid ed",
+ "▁re v",
+ "▁r ev",
+ "▁ rev",
+ "▁m ent",
+ "▁me nt",
+ "▁men t",
+ "▁ ment",
+ "an i",
+ "a ni",
+ "▁st ru",
+ "▁str u",
+ "▁ stru",
+ "▁att ention",
+ "art ment",
+ "▁I tal",
+ "▁It al",
+ "al le",
+ "all e",
+ "a lle",
+ "▁b is",
+ "▁bi s",
+ "▁ bis",
+ "ge ner",
+ "gen er",
+ "g ener",
+ "▁in tegr",
+ "▁int egr",
+ "▁inte gr",
+ "▁ integr",
+ "el lo",
+ "ell o",
+ "ry pt",
+ "▁a chie",
+ "ne s",
+ "n es",
+ "▁s tra",
+ "▁st ra",
+ "▁str a",
+ "▁ stra",
+ "s b",
+ "▁t ypes",
+ "▁type s",
+ "▁typ es",
+ "▁ty pes",
+ "▁ types",
+ "▁R E",
+ "▁ RE",
+ "In it",
+ "I nit",
+ "▁com ment",
+ "▁comm ent",
+ "▁comme nt",
+ "▁ comment",
+ "▁add ition",
+ "▁I D",
+ "▁ ID",
+ "AR T",
+ "A RT",
+ "F O",
+ "щ и",
+ "Con ne",
+ "Conn e",
+ "C onne",
+ "▁s qu",
+ "▁sq u",
+ "▁consider ed",
+ "▁consid ered",
+ "id ad",
+ "ida d",
+ "▁Oct ober",
+ "ci al",
+ "cia l",
+ "c ial",
+ "▁O f",
+ "▁ Of",
+ "▁tr avel",
+ "▁tra vel",
+ "▁trav el",
+ "▁b oy",
+ "▁bo y",
+ "▁ boy",
+ "') .",
+ "' ).",
+ "u y",
+ "il la",
+ "ill a",
+ "i lla",
+ "is try",
+ "ist ry",
+ "istr y",
+ "▁v a",
+ "▁ va",
+ "▁C he",
+ "▁Ch e",
+ "▁ Che",
+ "ER T",
+ "E RT",
+ "en de",
+ "end e",
+ "e nde",
+ "un gen",
+ "ung en",
+ "unge n",
+ "ab y",
+ "a by",
+ "▁R ober",
+ "▁Ro ber",
+ "▁Rob er",
+ "▁play ing",
+ "il s",
+ "i ls",
+ "▁s am",
+ "▁sa m",
+ "▁ sam",
+ "▁ex ecut",
+ "▁exec ut",
+ "▁ execut",
+ "▁U s",
+ "▁ Us",
+ "▁m ut",
+ "▁mu t",
+ "▁ mut",
+ "▁b al",
+ "▁ba l",
+ "▁ bal",
+ "as se",
+ "ass e",
+ "▁k ids",
+ "▁kid s",
+ "▁ki ds",
+ "▁fin anc",
+ "go r",
+ "g or",
+ "▁S ec",
+ "▁Se c",
+ "▁ Sec",
+ "ber t",
+ "be rt",
+ "b ert",
+ "▁H igh",
+ "▁Hig h",
+ "▁Hi gh",
+ "▁ High",
+ "▁ је",
+ "▁ke pt",
+ "but ton",
+ "b utton",
+ "it ory",
+ "itor y",
+ "ito ry",
+ "▁R em",
+ "▁Re m",
+ "▁ Rem",
+ "▁D E",
+ "▁ DE",
+ "▁re ach",
+ "▁r each",
+ "▁ reach",
+ "▁b ur",
+ "▁bu r",
+ "▁ bur",
+ "La bel",
+ "L abel",
+ "á t",
+ "ag o",
+ "a go",
+ "▁pass ed",
+ "▁pas sed",
+ "▁be hav",
+ "▁beh av",
+ "xF F",
+ "x FF",
+ "▁R eturn",
+ "▁Re turn",
+ "▁Ret urn",
+ "▁ Return",
+ "ST R",
+ "S TR",
+ "▁L es",
+ "▁Le s",
+ "▁ Les",
+ "▁o rd",
+ "▁or d",
+ "▁ ord",
+ "al a",
+ "a la",
+ "in ger",
+ "ing er",
+ "inge r",
+ "▁S ince",
+ "▁Sin ce",
+ "▁ Since",
+ "▁exper i",
+ "▁exp eri",
+ "▁s hall",
+ "▁sh all",
+ "▁sha ll",
+ "▁ shall",
+ "▁s tar",
+ "▁st ar",
+ "▁sta r",
+ "▁ star",
+ "no n",
+ "n on",
+ "▁g un",
+ "▁gu n",
+ "▁ gun",
+ "▁B el",
+ "▁Be l",
+ "▁ Bel",
+ "▁ob j",
+ "▁ obj",
+ "ar es",
+ "are s",
+ "a res",
+ "r s",
+ "▁we eks",
+ "▁week s",
+ "ne n",
+ "n en",
+ "▁S tre",
+ "▁St re",
+ "▁Str e",
+ "or ing",
+ "ori ng",
+ "o ring",
+ "▁ î",
+ "▁ser ious",
+ "time s",
+ "ti mes",
+ "tim es",
+ "t imes",
+ "▁H ouse",
+ "▁Ho use",
+ "▁Hou se",
+ "▁r oll",
+ "▁ro ll",
+ "▁ roll",
+ "▁reg ister",
+ "▁ register",
+ "▁mod ule",
+ "▁mo dule",
+ "▁ module",
+ "▁app lic",
+ "▁ap plic",
+ "▁appl ic",
+ "I R",
+ "▁c ook",
+ "▁co ok",
+ "▁ cook",
+ "au x",
+ "a ux",
+ "▁s ave",
+ "▁sa ve",
+ "▁sav e",
+ "▁ save",
+ "▁C r",
+ "▁ Cr",
+ ", \r",
+ "▁st ates",
+ "▁stat es",
+ "▁state s",
+ "▁sta tes",
+ "▁ states",
+ "▁em pty",
+ "▁emp ty",
+ "▁empt y",
+ "▁ empty",
+ "▁aut om",
+ "▁au tom",
+ "▁auto m",
+ "▁ autom",
+ "fig ure",
+ "ian ce",
+ "i ance",
+ "▁h appy",
+ "▁happ y",
+ "▁f n",
+ "▁ fn",
+ "▁j ud",
+ "▁ju d",
+ "▁ jud",
+ "▁h at",
+ "▁ha t",
+ "▁ hat",
+ "AC K",
+ "A CK",
+ "▁F e",
+ "▁ Fe",
+ "$ -",
+ "iv il",
+ "ivi l",
+ "i vil",
+ "ot ed",
+ "ote d",
+ "o ted",
+ "▁size of",
+ "▁ sizeof",
+ "▁sit uation",
+ "▁situ ation",
+ "▁l ives",
+ "▁li ves",
+ "▁live s",
+ "▁liv es",
+ "▁fe eling",
+ "▁feel ing",
+ "▁fee ling",
+ "▁r isk",
+ "▁ri sk",
+ "▁ris k",
+ "▁Jan uary",
+ "▁Januar y",
+ "▁Ob ject",
+ "▁ Object",
+ "▁re comm",
+ "▁rec omm",
+ "▁в ы",
+ "▁ вы",
+ "▁pot ential",
+ "ea h",
+ "e ah",
+ "▁com plex",
+ "▁comp lex",
+ "▁compl ex",
+ "▁ complex",
+ "print f",
+ "ist ance",
+ "istan ce",
+ "i stance",
+ "ir th",
+ "irt h",
+ "li k",
+ "l ik",
+ "as te",
+ "ast e",
+ "a ste",
+ "▁wh ose",
+ "▁who se",
+ "Ar g",
+ "A rg",
+ "▁mod ern",
+ "▁mo dern",
+ "▁mode rn",
+ "▁moder n",
+ "ion es",
+ "io nes",
+ "ione s",
+ "i ones",
+ "▁ч е",
+ "▁ че",
+ "▁s ett",
+ "▁se tt",
+ "▁set t",
+ "▁M ag",
+ "▁Ma g",
+ "▁ Mag",
+ "a e",
+ "▁cond ition",
+ "▁ condition",
+ "Le ngth",
+ "L ength",
+ "▁f it",
+ "▁fi t",
+ "▁ fit",
+ "ound s",
+ "oun ds",
+ "▁ch anged",
+ "▁chang ed",
+ "▁change d",
+ "▁ changed",
+ "▁g uy",
+ "▁gu y",
+ "fil ter",
+ "at ever",
+ "ate ver",
+ "é d",
+ "re move",
+ "rem ove",
+ "▁h op",
+ "▁ho p",
+ "▁ hop",
+ "▁O ut",
+ "▁ Out",
+ "▁R ich",
+ "▁Ric h",
+ "▁ Rich",
+ "ch ild",
+ "chi ld",
+ "▁in cluded",
+ "▁incl uded",
+ "▁includ ed",
+ "▁include d",
+ "▁inclu ded",
+ "$ \\",
+ "▁T om",
+ "▁To m",
+ "▁ Tom",
+ "el ine",
+ "eli ne",
+ "elin e",
+ "e line",
+ "▁s ometimes",
+ "▁some times",
+ "▁somet imes",
+ "▁sometime s",
+ "▁dr ink",
+ "▁qu ant",
+ "▁ quant",
+ "▁p lease",
+ "▁ple ase",
+ "▁I nt",
+ "▁In t",
+ "▁ Int",
+ "ri ef",
+ "rie f",
+ "r ief",
+ "▁ex actly",
+ "▁exact ly",
+ "ci ng",
+ "cin g",
+ "c ing",
+ "▁all owed",
+ "▁allow ed",
+ "▁ allowed",
+ "bu ild",
+ "b uild",
+ "▁beaut iful",
+ "▁W ell",
+ "▁We ll",
+ "▁Wel l",
+ "▁ Well",
+ "▁look s",
+ "▁lo oks",
+ "▁ ü",
+ "▁ch ance",
+ "▁w rote",
+ "▁wr ote",
+ "▁n or",
+ "▁no r",
+ "▁ nor",
+ "▁f ailed",
+ "▁fa iled",
+ "▁fail ed",
+ "▁ failed",
+ "Me t",
+ "M et",
+ "▁p rior",
+ "▁pr ior",
+ "▁pri or",
+ "▁h undred",
+ "ско й",
+ "с кой",
+ "or ia",
+ "ori a",
+ "o ria",
+ "▁c y",
+ "▁ cy",
+ "▁w eb",
+ "▁we b",
+ "▁ web",
+ "▁m ess",
+ "▁me ss",
+ "▁mes s",
+ "le q",
+ "l eq",
+ "d y",
+ "te x",
+ "t ex",
+ "▁a nim",
+ "▁an im",
+ "▁ anim",
+ "at ur",
+ "atu r",
+ "▁str ucture",
+ "▁struct ure",
+ "▁ structure",
+ "opt ion",
+ "o ption",
+ "▁act ual",
+ "▁ actual",
+ "▁Fr anc",
+ "▁Fra nc",
+ "▁Fran c",
+ "en ced",
+ "ence d",
+ "enc ed",
+ ".< /",
+ ". ",
+ "▁f low",
+ "▁fl ow",
+ "▁flo w",
+ "▁ flow",
+ "▁A fr",
+ "▁Af r",
+ "de t",
+ "d et",
+ "▁K e",
+ "▁ Ke",
+ "et y",
+ "e ty",
+ "ски й",
+ "с кий",
+ "▁st uff",
+ "it ter",
+ "itt er",
+ "itte r",
+ "▁ar gs",
+ "▁arg s",
+ "▁ args",
+ "▁al bum",
+ "▁ album",
+ "▁ ]",
+ "ug in",
+ "u gin",
+ "S U",
+ "Pe r",
+ "P er",
+ "▁cir c",
+ "▁ci rc",
+ "▁ circ",
+ "▁cor rect",
+ "▁corre ct",
+ "▁ correct",
+ "▁l ines",
+ "▁li nes",
+ "▁line s",
+ "▁lin es",
+ "▁ lines",
+ "▁complet ely",
+ "▁complete ly",
+ "kn own",
+ "know n",
+ "k nown",
+ "▁t ree",
+ "▁tr ee",
+ "▁tre e",
+ "▁ tree",
+ "ro ot",
+ "r oot",
+ "▁J apan",
+ "▁Ja pan",
+ "▁Jap an",
+ "ol es",
+ "ole s",
+ "o les",
+ "en do",
+ "end o",
+ "▁l ocation",
+ "▁loc ation",
+ "▁ location",
+ "▁ Х",
+ "▁m id",
+ "▁mi d",
+ "▁ mid",
+ "al ing",
+ "ali ng",
+ "alin g",
+ "a ling",
+ "G L",
+ "ia no",
+ "ian o",
+ "i ano",
+ "▁{ }",
+ "▁ {}",
+ "la ng",
+ "lan g",
+ "l ang",
+ "▁equ ip",
+ "ERR OR",
+ "▁mem ory",
+ "▁memor y",
+ "▁memo ry",
+ "▁ memory",
+ "▁( \"",
+ "▁ (\"",
+ "▁n ature",
+ "▁nat ure",
+ "▁natur e",
+ "go ogle",
+ "ab s",
+ "a bs",
+ "B C",
+ "▁g ets",
+ "▁get s",
+ "▁ge ts",
+ "▁ gets",
+ "Com mand",
+ "Comm and",
+ "TE R",
+ "T ER",
+ "al ed",
+ "ale d",
+ "a led",
+ "c p",
+ "▁p urch",
+ "▁pur ch",
+ "▁D en",
+ "▁De n",
+ "▁ Den",
+ "▁her self",
+ "▁hers elf",
+ "▁I r",
+ "▁ Ir",
+ "▁s ie",
+ "▁si e",
+ "ga r",
+ "g ar",
+ "A p",
+ "▁n el",
+ "▁ne l",
+ "▁ nel",
+ "ot a",
+ "o ta",
+ ") ]",
+ "co r",
+ "c or",
+ "ac ht",
+ "ach t",
+ "a cht",
+ "( *",
+ "irt ual",
+ "▁pol ice",
+ "▁polic e",
+ "▁s kin",
+ "▁sk in",
+ "▁ski n",
+ "▁ skin",
+ "sh ip",
+ "s hip",
+ "ef ined",
+ "augh ter",
+ "aught er",
+ "in ding",
+ "ind ing",
+ "indi ng",
+ "▁S l",
+ "▁ Sl",
+ "▁in flu",
+ "▁infl u",
+ "▁inf lu",
+ "▁m ount",
+ "▁mo unt",
+ "▁mou nt",
+ "▁ mount",
+ "▁a z",
+ "▁ az",
+ "▁w ood",
+ "▁wo od",
+ "▁ wood",
+ "ot es",
+ "ote s",
+ "o tes",
+ "eg a",
+ "e ga",
+ "▁acc ording",
+ "▁accord ing",
+ "▁name space",
+ "▁names pace",
+ "▁ namespace",
+ "Del ta",
+ "D elta",
+ "st ant",
+ "sta nt",
+ "stan t",
+ "▁pub lished",
+ "▁publish ed",
+ "▁ published",
+ "ak er",
+ "ake r",
+ "a ker",
+ "▁Bl ack",
+ "▁ Black",
+ "l n",
+ "▁indust ry",
+ "SO N",
+ "S ON",
+ "Re p",
+ "R ep",
+ "▁ch oice",
+ "▁cho ice",
+ "▁ choice",
+ "▁in n",
+ "▁i nn",
+ "▁ inn",
+ "k l",
+ "▁p al",
+ "▁pa l",
+ "▁ pal",
+ "▁a ud",
+ "▁au d",
+ "▁ aud",
+ "▁stand ard",
+ "▁ standard",
+ "▁know ledge",
+ "** ,",
+ "* *,",
+ "▁F rank",
+ "▁Fr ank",
+ "▁Fran k",
+ "s q",
+ "Out put",
+ "▁f ör",
+ "▁fö r",
+ "▁ för",
+ "Val id",
+ "ug h",
+ "u gh",
+ "▁bo oks",
+ "▁book s",
+ "▁ books",
+ "▁J ames",
+ "▁Jam es",
+ "▁Ja mes",
+ "k o",
+ "▁compan ies",
+ "an ning",
+ "ann ing",
+ "anni ng",
+ "▁v ict",
+ "▁vi ct",
+ "▁vic t",
+ "▁re pl",
+ "▁rep l",
+ "▁s che",
+ "▁sc he",
+ "▁sch e",
+ "▁ sche",
+ "▁h appen",
+ "▁happ en",
+ "▁ha ppen",
+ "ft y",
+ "f ty",
+ "ac ity",
+ "aci ty",
+ "a city",
+ "ir a",
+ "i ra",
+ "▁im plement",
+ "▁imp lement",
+ "▁impl ement",
+ "▁ implement",
+ "ско го",
+ "ск ого",
+ "с кого",
+ "num ber",
+ "nu mber",
+ "n umber",
+ "S H",
+ "ir o",
+ "i ro",
+ "▁f ear",
+ "▁fe ar",
+ "▁t ouch",
+ "▁to uch",
+ "▁tou ch",
+ "▁ touch",
+ "▁c ast",
+ "▁cas t",
+ "▁ca st",
+ "▁ cast",
+ "AS S",
+ "A SS",
+ "▁cons ist",
+ "T ask",
+ "▁s ig",
+ "▁si g",
+ "▁ sig",
+ "б а",
+ "ig ation",
+ "▁M ost",
+ "▁Mo st",
+ "▁Mos t",
+ "▁ Most",
+ "▁D er",
+ "▁De r",
+ "▁ Der",
+ "}( \\",
+ "} (\\",
+ ": \"",
+ "▁F ig",
+ "▁Fi g",
+ "▁ Fig",
+ "al i",
+ "a li",
+ "in er",
+ "ine r",
+ "i ner",
+ "') ,",
+ "' ),",
+ "▁C oun",
+ "▁Co un",
+ "▁Cou n",
+ "( _",
+ "▁d istributed",
+ "▁distribut ed",
+ "▁distribute d",
+ "NA ME",
+ "N AME",
+ "▁m ur",
+ "▁mu r",
+ "▁care er",
+ "~ ~",
+ "pe rs",
+ "per s",
+ "p ers",
+ "ar ies",
+ "ari es",
+ "a ries",
+ "en ses",
+ "ens es",
+ "ense s",
+ "▁Al so",
+ "▁Als o",
+ "Vers ion",
+ "V ersion",
+ "▁un ique",
+ "▁uniqu e",
+ "▁ unique",
+ "▁Fr ance",
+ "▁Franc e",
+ "▁Fran ce",
+ "B A",
+ "k y",
+ "▁F ebru",
+ "▁Fe bru",
+ "▁Feb ru",
+ "▁d ied",
+ "▁di ed",
+ "▁die d",
+ "om ega",
+ "ome ga",
+ "▁F orm",
+ "▁For m",
+ "▁Fo rm",
+ "▁ Form",
+ "▁w idth",
+ "▁wid th",
+ "▁ width",
+ "to col",
+ "t ocol",
+ "▁l ie",
+ "▁li e",
+ "▁ lie",
+ "Sh e",
+ "S he",
+ "é m",
+ "▁stra ight",
+ "▁n ach",
+ "▁na ch",
+ "▁st ood",
+ "▁sto od",
+ "▁ stood",
+ "ol ds",
+ "old s",
+ "▁g oes",
+ "▁go es",
+ "ce ll",
+ "cel l",
+ "c ell",
+ "▁t ill",
+ "▁til l",
+ "▁ti ll",
+ "L I",
+ "dr aw",
+ "d raw",
+ "▁s atisf",
+ "▁sat isf",
+ "▁re ading",
+ "▁read ing",
+ "AT ION",
+ "A TION",
+ "▁A re",
+ "▁Ar e",
+ "▁ Are",
+ "▁A c",
+ "▁ Ac",
+ ") *",
+ "▁add itional",
+ "▁addition al",
+ "wo od",
+ "w ood",
+ "ci l",
+ "c il",
+ "п у",
+ "UL T",
+ "U LT",
+ "▁b ill",
+ "▁bi ll",
+ "▁bil l",
+ "ma s",
+ "m as",
+ "an ia",
+ "ani a",
+ "a nia",
+ "с у",
+ "an z",
+ "he ight",
+ "h eight",
+ "j o",
+ "▁d os",
+ "▁do s",
+ "\\ \"",
+ "▁/ >",
+ "▁ />",
+ "▁p roduction",
+ "▁produ ction",
+ "▁product ion",
+ "▁prod uction",
+ "▁ production",
+ "ig er",
+ "ige r",
+ "i ger",
+ "▁с т",
+ "▁ ст",
+ "sh ow",
+ "s how",
+ "▁pop ulation",
+ "▁popul ation",
+ "▁p ark",
+ "▁par k",
+ "▁ park",
+ "▁Z e",
+ "▁necess ary",
+ "▁ necessary",
+ "▁t rust",
+ "▁tr ust",
+ "▁sh own",
+ "▁show n",
+ "mod ule",
+ "mo dule",
+ "G E",
+ "▁l ay",
+ "▁la y",
+ "▁ lay",
+ "▁ann oun",
+ "▁class Name",
+ "▁ className",
+ "▁cal cul",
+ "▁calc ul",
+ "Fun ction",
+ "F unction",
+ "▁S al",
+ "▁Sa l",
+ "▁ Sal",
+ "O K",
+ "T P",
+ "▁en try",
+ "▁ent ry",
+ "▁entr y",
+ "▁ entry",
+ "▁St ud",
+ "▁ Stud",
+ "▁it ems",
+ "▁item s",
+ "▁ items",
+ "▁se curity",
+ "▁sec urity",
+ "▁secur ity",
+ "▁ security",
+ "En try",
+ "Ent ry",
+ "f loat",
+ "l s",
+ "ib ly",
+ "▁cont ribut",
+ "▁C heck",
+ "▁Che ck",
+ "▁ Check",
+ "M D",
+ "▁impro ve",
+ "Par t",
+ "P art",
+ "▁system s",
+ "▁syst ems",
+ "B l",
+ "▁pol icy",
+ "▁polic y",
+ "▁ policy",
+ "▁s creen",
+ "▁sc reen",
+ "▁scr een",
+ "▁ screen",
+ "▁A ny",
+ "▁An y",
+ "▁ Any",
+ "▁op ened",
+ "▁open ed",
+ "al loc",
+ "all oc",
+ "allo c",
+ "▁De cember",
+ "▁Dec ember",
+ "▁ É",
+ "▁e mail",
+ "▁em ail",
+ "▁ email",
+ "ad er",
+ "ade r",
+ "a der",
+ "= >",
+ "▁H en",
+ "▁He n",
+ "▁ Hen",
+ "▁in fo",
+ "▁inf o",
+ "▁ info",
+ "▁f loat",
+ "▁flo at",
+ "▁ float",
+ "▁sw itch",
+ "▁ switch",
+ "ра н",
+ "р ан",
+ "ur ance",
+ "▁as sum",
+ "▁ass um",
+ "us tr",
+ "ust r",
+ "u str",
+ "▁g roups",
+ "▁group s",
+ "▁gro ups",
+ "▁ groups",
+ "▁R ead",
+ "▁Re ad",
+ "▁ Read",
+ "▁w at",
+ "▁wa t",
+ "S p",
+ "ве р",
+ "в ер",
+ "RA N",
+ "R AN",
+ "hi b",
+ "h ib",
+ "AL L",
+ "A LL",
+ "▁h us",
+ "▁ hus",
+ "Sp ec",
+ "Spe c",
+ "S pec",
+ "\") )",
+ "\" ))",
+ "▁F rench",
+ "▁C lass",
+ "▁Cl ass",
+ "▁ Class",
+ "▁pres ident",
+ "▁presid ent",
+ "▁def init",
+ "▁defin it",
+ "▁N or",
+ "▁No r",
+ "▁T hom",
+ "▁Th om",
+ "ai gn",
+ "a ign",
+ "W idth",
+ "D o",
+ "▁{ @",
+ "ag on",
+ "ago n",
+ "a gon",
+ "▁L u",
+ "▁ Lu",
+ "▁follow ed",
+ "M M",
+ "as ons",
+ "ason s",
+ "tm p",
+ "t mp",
+ "▁th rows",
+ "▁throw s",
+ "▁thr ows",
+ "▁thro ws",
+ "▁ throws",
+ "IT Y",
+ "I TY",
+ "но м",
+ "▁f air",
+ "▁fa ir",
+ "▁p en",
+ "▁pe n",
+ "▁ pen",
+ "é g",
+ "▁inter face",
+ "▁ interface",
+ "▁s af",
+ "▁sa f",
+ "oo n",
+ "o on",
+ "B ack",
+ "▁s peed",
+ "▁sp eed",
+ "▁spe ed",
+ "▁ speed",
+ "▁ext ends",
+ "▁extend s",
+ "em pty",
+ "empt y",
+ "emp ty",
+ "▁п ере",
+ "▁пер е",
+ "▁пе ре",
+ "▁pro per",
+ "▁pr oper",
+ "▁prop er",
+ "▁d riv",
+ "▁dr iv",
+ "▁dri v",
+ "ф и",
+ "▁c enter",
+ "▁cent er",
+ "▁ center",
+ "he ader",
+ "head er",
+ "▁} )",
+ "▁ })",
+ "w a",
+ "▁m iddle",
+ "▁ middle",
+ "▁ch oose",
+ "▁cho ose",
+ "▁St ad",
+ "▁Sta d",
+ "S O",
+ "Fact ory",
+ "Factor y",
+ "F actory",
+ "De v",
+ "D ev",
+ "ic les",
+ "icle s",
+ "icl es",
+ "i cles",
+ "▁ap plication",
+ "▁applic ation",
+ "▁appl ication",
+ "▁ application",
+ "▁mod els",
+ "▁model s",
+ "▁mode ls",
+ "▁ models",
+ "pi te",
+ "pit e",
+ "p ite",
+ "ca p",
+ "c ap",
+ "x i",
+ "osp ital",
+ "▁d ream",
+ "▁dre am",
+ "EN D",
+ "E ND",
+ "▁con tract",
+ "▁cont ract",
+ "▁contr act",
+ "▁contra ct",
+ "▁ contract",
+ "icro soft",
+ "▁th ous",
+ "▁thou s",
+ "iz es",
+ "ize s",
+ "i zes",
+ "▁д а",
+ "▁ да",
+ "▁C O",
+ "▁ CO",
+ "▁d irection",
+ "▁di rection",
+ "▁direct ion",
+ "▁dire ction",
+ "▁dir ection",
+ "▁ direction",
+ "▁` `",
+ "▁ ``",
+ "▁d rive",
+ "▁dr ive",
+ "▁dri ve",
+ "▁driv e",
+ "▁ drive",
+ "Ma x",
+ "M ax",
+ "ci a",
+ "c ia",
+ "▁contin u",
+ "▁A lex",
+ "▁Al ex",
+ "▁Ale x",
+ "▁ Alex",
+ "▁g old",
+ "▁go ld",
+ "▁gol d",
+ "▁ gold",
+ "▁p rep",
+ "▁pre p",
+ "▁pr ep",
+ "▁or igin",
+ "▁orig in",
+ "▁ origin",
+ "▁r ap",
+ "▁ra p",
+ "▁ rap",
+ "O p",
+ "ous ly",
+ "▁are as",
+ "▁area s",
+ "PO RT",
+ "P ORT",
+ "он а",
+ "о на",
+ "▁sa fe",
+ "▁saf e",
+ "▁ safe",
+ "▁profess ional",
+ "▁profession al",
+ "ap ache",
+ "apa che",
+ "▁t emper",
+ "▁tem per",
+ "▁temp er",
+ "s z",
+ "▁u nit",
+ "▁un it",
+ "▁ unit",
+ "▁c op",
+ "▁co p",
+ "▁ cop",
+ "eq n",
+ "List ener",
+ "Listen er",
+ "▁for mat",
+ "▁form at",
+ "▁forma t",
+ "▁ format",
+ "se lect",
+ "sel ect",
+ "s elect",
+ "▁com fort",
+ "▁ comfort",
+ "▁me ant",
+ "▁mean t",
+ "id ay",
+ "ida y",
+ "i day",
+ "em e",
+ "e me",
+ "▁act ive",
+ "▁activ e",
+ "▁ active",
+ "▁n ote",
+ "▁not e",
+ "▁no te",
+ "▁ note",
+ "▁M il",
+ "▁Mi l",
+ "▁ Mil",
+ "on ly",
+ "▁< =",
+ "▁ <=",
+ "▁ne igh",
+ "▁nei gh",
+ "a o",
+ "▁bl ue",
+ "▁ blue",
+ "▁T V",
+ "▁ TV",
+ "Ch ild",
+ "▁re ached",
+ "▁reach ed",
+ "Add ress",
+ "Addr ess",
+ "ст в",
+ "▁cl osed",
+ "▁close d",
+ "▁clos ed",
+ "▁clo sed",
+ "▁ closed",
+ "in der",
+ "ind er",
+ "inde r",
+ "i nder",
+ "ol o",
+ "o lo",
+ "▁a lt",
+ "▁al t",
+ "▁ alt",
+ "▁a dm",
+ "▁ad m",
+ "Form at",
+ "For mat",
+ "U I",
+ "▁H am",
+ "▁Ha m",
+ "▁f requ",
+ "▁fr equ",
+ "▁fre qu",
+ "▁in depend",
+ "▁inde pend",
+ "▁ independ",
+ "▁eas ily",
+ "▁L and",
+ "▁La nd",
+ "▁Lan d",
+ "▁ Land",
+ "▁t or",
+ "▁to r",
+ "▁ tor",
+ "ograph y",
+ "ograp hy",
+ "in fty",
+ "inf ty",
+ "▁W ork",
+ "▁Wor k",
+ "▁ Work",
+ "iv en",
+ "ive n",
+ "i ven",
+ "▁Count y",
+ "▁Coun ty",
+ "▁s rc",
+ "▁ src",
+ "}$ ,",
+ "} $,",
+ "par se",
+ "pars e",
+ "p arse",
+ "C D",
+ "▁C our",
+ "▁Co ur",
+ "▁Cou r",
+ "▁f ol",
+ "▁fo l",
+ "▁ fol",
+ "Ent ity",
+ "pg f",
+ "▁Ch ina",
+ "▁Chi na",
+ "▁S ub",
+ "▁Su b",
+ "▁ Sub",
+ "ho od",
+ "h ood",
+ "▁field s",
+ "▁ fields",
+ "▁y es",
+ "▁ye s",
+ "▁ yes",
+ "re nd",
+ "ren d",
+ "r end",
+ "▁to wards",
+ "▁toward s",
+ "▁tow ards",
+ "▁st aff",
+ "▁sta ff",
+ "▁ staff",
+ "▁A ir",
+ "▁ Air",
+ "▁st ation",
+ "▁stat ion",
+ "▁ station",
+ "at ives",
+ "ative s",
+ "ati ves",
+ "ativ es",
+ "▁imp act",
+ "в ы",
+ "▁direct ly",
+ "iss ions",
+ "ission s",
+ "iv a",
+ "i va",
+ "| \\",
+ "Pt r",
+ "P tr",
+ "▁S ant",
+ "▁San t",
+ "▁Sa nt",
+ "Po l",
+ "P ol",
+ "▁pro gress",
+ "▁ progress",
+ "it ar",
+ "ita r",
+ "i tar",
+ "▁p arts",
+ "▁part s",
+ "▁par ts",
+ "▁ parts",
+ "▁pl ant",
+ "▁plan t",
+ "▁ plant",
+ "▁abs olut",
+ "▁gu ess",
+ "eq ref",
+ "▁t im",
+ "▁ti m",
+ "▁ tim",
+ "▁L ou",
+ "▁Lo u",
+ "▁ Lou",
+ "▁c ool",
+ "▁co ol",
+ "al u",
+ "a lu",
+ "▁m outh",
+ "▁mo uth",
+ "▁mou th",
+ "▁ mouth",
+ "ни х",
+ "▁h eight",
+ "▁he ight",
+ "▁ height",
+ "ge st",
+ "ges t",
+ "g est",
+ "▁P ost",
+ "▁Po st",
+ "▁Pos t",
+ "▁ Post",
+ "▁b oard",
+ "▁bo ard",
+ "▁ board",
+ "▁t it",
+ "▁ti t",
+ "▁ tit",
+ "▁h our",
+ "▁ho ur",
+ "▁ hour",
+ "▁ser ver",
+ "▁serv er",
+ "▁serve r",
+ "▁ server",
+ "▁p layers",
+ "▁play ers",
+ "▁player s",
+ "ri er",
+ "rie r",
+ "r ier",
+ "Lin k",
+ "L ink",
+ "▁Pres ident",
+ "] (",
+ "▁con struct",
+ "▁const ruct",
+ "▁constr uct",
+ "▁constru ct",
+ "▁ construct",
+ "hand le",
+ "}$ .",
+ "} $.",
+ "ry ing",
+ "r ying",
+ "▁s hop",
+ "▁sh op",
+ "▁ shop",
+ "ia na",
+ "ian a",
+ "i ana",
+ "ex p",
+ "e xp",
+ "Hel per",
+ "Help er",
+ "Off set",
+ "ac hes",
+ "ach es",
+ "ache s",
+ "a ches",
+ "▁conne ction",
+ "▁connect ion",
+ "▁conn ection",
+ "▁ connection",
+ "▁d ifference",
+ "▁dif ference",
+ "▁differ ence",
+ "serv ice",
+ "s ervice",
+ "▁g as",
+ "▁ga s",
+ "▁ gas",
+ "▁p riv",
+ "▁pr iv",
+ "▁pri v",
+ "▁ priv",
+ "▁un ivers",
+ "▁ univers",
+ "▁w ish",
+ "▁wis h",
+ "Re m",
+ "R em",
+ "U rl",
+ "ge b",
+ "g eb",
+ "S o",
+ "ens ions",
+ "ension s",
+ "Mod ule",
+ "Mo dule",
+ "SI ZE",
+ "▁p rem",
+ "▁pre m",
+ "▁pr em",
+ "wind ow",
+ "w indow",
+ "▁d ies",
+ "▁di es",
+ "▁die s",
+ "de l",
+ "d el",
+ "▁r ow",
+ "▁ro w",
+ "▁ row",
+ "▁a verage",
+ "▁aver age",
+ "▁ave rage",
+ "xi m",
+ "x im",
+ "▁p u",
+ "▁ pu",
+ "an ç",
+ "De t",
+ "D et",
+ "ke r",
+ "k er",
+ "y a",
+ "▁D et",
+ "▁De t",
+ "▁ Det",
+ "▁p å",
+ "▁n amed",
+ "▁name d",
+ "▁na med",
+ "▁nam ed",
+ "▁ named",
+ "▁dec ision",
+ "▁decis ion",
+ "wi n",
+ "w in",
+ "▁Ge orge",
+ "▁Georg e",
+ "ar ily",
+ "ari ly",
+ "▁s olution",
+ "▁sol ution",
+ "▁mult iple",
+ "▁multi ple",
+ "▁multip le",
+ "▁ multiple",
+ "at egy",
+ "ate gy",
+ "ateg y",
+ "▁le arning",
+ "▁learn ing",
+ "▁lear ning",
+ "▁ learning",
+ "▁se cret",
+ "▁sec ret",
+ "▁secre t",
+ "▁ secret",
+ "D O",
+ "▁n ice",
+ "▁ni ce",
+ "▁nic e",
+ "▁ nice",
+ "//////// ////////",
+ "S u",
+ "it ation",
+ "itat ion",
+ "▁j oin",
+ "▁jo in",
+ "▁ join",
+ "▁el ements",
+ "▁element s",
+ "▁ele ments",
+ "▁elem ents",
+ "▁ elements",
+ "▁e mer",
+ "▁em er",
+ "til de",
+ "t ilde",
+ "▁d ep",
+ "▁de p",
+ "▁ dep",
+ "▁s hot",
+ "▁sh ot",
+ "▁ shot",
+ "▁pl atform",
+ "▁plat form",
+ "▁ platform",
+ "ot hing",
+ "oth ing",
+ "o thing",
+ "M y",
+ "ed ia",
+ "edi a",
+ "om s",
+ "o ms",
+ "ail y",
+ "ai ly",
+ "a ily",
+ "( [",
+ "▁d ress",
+ "▁dr ess",
+ "▁dre ss",
+ "▁off icial",
+ "▁offic ial",
+ "es tern",
+ "est ern",
+ "ester n",
+ "este rn",
+ "▁dis cover",
+ "▁disc over",
+ "▁disco ver",
+ "▁m i",
+ "▁ mi",
+ "ны е",
+ "C A",
+ "od ing",
+ "odi ng",
+ "o ding",
+ "▁F ound",
+ "▁Fou nd",
+ "▁Fo und",
+ "▁ Found",
+ "▁a ffect",
+ "▁aff ect",
+ "▁af fect",
+ "Vi s",
+ "V is",
+ "st ract",
+ "str act",
+ "stra ct",
+ "s tract",
+ "ic ed",
+ "ice d",
+ "i ced",
+ "de bug",
+ "d ebug",
+ "▁rel ated",
+ "▁relate d",
+ "▁ related",
+ "▁s pect",
+ "▁sp ect",
+ "▁spec t",
+ "▁spe ct",
+ "▁ spect",
+ "us hed",
+ "ush ed",
+ "сь ко",
+ "▁b ank",
+ "▁ban k",
+ "▁ bank",
+ "▁c ele",
+ "▁ce le",
+ "▁cel e",
+ "AN D",
+ "A ND",
+ "ol f",
+ "е м",
+ "▁f ill",
+ "▁fil l",
+ "▁fi ll",
+ "▁ fill",
+ "▁g ives",
+ "▁giv es",
+ "▁give s",
+ "▁gi ves",
+ "▁б у",
+ "▁ бу",
+ "ar on",
+ "aro n",
+ "a ron",
+ "▁J es",
+ "▁Je s",
+ "RE G",
+ "▁s udd",
+ "▁su dd",
+ "▁sud d",
+ "date d",
+ "da ted",
+ "dat ed",
+ "d ated",
+ "v i",
+ "▁g i",
+ "▁ gi",
+ "se nd",
+ "sen d",
+ "s end",
+ "cp p",
+ "c pp",
+ "▁s pent",
+ "▁sp ent",
+ "▁spe nt",
+ "an de",
+ "and e",
+ "a nde",
+ "▁oper ation",
+ "▁ operation",
+ "pro cess",
+ "proc ess",
+ "▁in form",
+ "▁inf orm",
+ "▁info rm",
+ "▁F ree",
+ "▁Fr ee",
+ "▁Fre e",
+ "▁ Free",
+ "yo nd",
+ "y ond",
+ "▁per haps",
+ "▁su rv",
+ "▁sur v",
+ "▁L oc",
+ "▁Lo c",
+ "▁ Loc",
+ "▁con cl",
+ "▁conc l",
+ "▁ра з",
+ "▁ раз",
+ "▁O ver",
+ "▁ Over",
+ "ho l",
+ "h ol",
+ "ra z",
+ "r az",
+ "Wr ite",
+ "Writ e",
+ "W rite",
+ "▁g iving",
+ "▁giv ing",
+ "▁gi ving",
+ "r d",
+ "in stance",
+ "inst ance",
+ "▁re leased",
+ "▁rele ased",
+ "▁release d",
+ "▁R o",
+ "▁ Ro",
+ "R A",
+ "▁pract ice",
+ "▁g raph",
+ "▁gr aph",
+ "▁gra ph",
+ "▁grap h",
+ "▁ graph",
+ "▁incre ase",
+ "▁fig ure",
+ "▁ figure",
+ "Fil ter",
+ "HE CK",
+ "id x",
+ "i dx",
+ "▁g lass",
+ "▁gl ass",
+ "▁ glass",
+ "sk i",
+ "s ki",
+ "com es",
+ "co mes",
+ "come s",
+ "c omes",
+ "▁c at",
+ "▁ca t",
+ "▁ cat",
+ "▁c old",
+ "▁col d",
+ "▁co ld",
+ "go to",
+ "got o",
+ "g oto",
+ "uf act",
+ "u fact",
+ "▁C opyright",
+ "▁Copy right",
+ "▁ Copyright",
+ "}} \\",
+ "} }\\",
+ "▁str eng",
+ "▁stre ng",
+ "▁d ir",
+ "▁di r",
+ "▁ dir",
+ "to ken",
+ "tok en",
+ "t oken",
+ "▁occ ur",
+ "▁oc cur",
+ "arl ier",
+ "▁me asure",
+ "▁meas ure",
+ "▁ measure",
+ "▁s ec",
+ "▁se c",
+ "▁ sec",
+ "▁m ás",
+ "▁má s",
+ "▁N et",
+ "▁Ne t",
+ "▁ Net",
+ "▁arg ument",
+ "▁ argument",
+ "▁s ou",
+ "▁so u",
+ "▁m oving",
+ "▁mov ing",
+ "▁mo ving",
+ "▁p refer",
+ "▁pre fer",
+ "▁pref er",
+ "ma sk",
+ "mas k",
+ "m ask",
+ "< <",
+ "▁bre ath",
+ "▁breat h",
+ "▁phys ical",
+ "▁pos itive",
+ "▁posit ive",
+ "▁s or",
+ "▁so r",
+ "▁ sor",
+ "▁de part",
+ "▁dep art",
+ "▁re move",
+ "▁rem ove",
+ "▁ remove",
+ "▁k it",
+ "▁ki t",
+ "▁ kit",
+ "▁me eting",
+ "▁meet ing",
+ "▁D ata",
+ "▁Da ta",
+ "▁Dat a",
+ "▁ Data",
+ "og raf",
+ "act ions",
+ "action s",
+ "a ctions",
+ "▁param eters",
+ "▁parameter s",
+ "▁ parameters",
+ "▁A tt",
+ "▁At t",
+ "▁ Att",
+ "es ch",
+ "esc h",
+ "e sch",
+ "▁inv olved",
+ "▁invol ved",
+ "▁involve d",
+ "ä t",
+ "L L",
+ "B ar",
+ "▁с и",
+ "▁ си",
+ "ec h",
+ "e ch",
+ "GE T",
+ "G ET",
+ "▁pre vent",
+ "▁pr event",
+ "▁prev ent",
+ "▁ prevent",
+ "▁be yond",
+ "▁O ther",
+ "▁Ot her",
+ "▁ Other",
+ "ä n",
+ "by te",
+ "▁sudd en",
+ "▁sud den",
+ "ol ve",
+ "olv e",
+ "▁н о",
+ "▁ но",
+ "LO G",
+ "L OG",
+ "un it",
+ "uni t",
+ "u nit",
+ "▁tr uth",
+ "ra t",
+ "r at",
+ "S D",
+ "▁e at",
+ "▁M ad",
+ "▁Ma d",
+ "▁ Mad",
+ "▁prov ides",
+ "▁provide s",
+ "▁s ession",
+ "▁ session",
+ "De le",
+ "Del e",
+ "D ele",
+ "▁con vers",
+ "▁conv ers",
+ "▁conver s",
+ "▁conve rs",
+ "cent er",
+ "cen ter",
+ "c enter",
+ "▁contin ued",
+ "▁continue d",
+ "▁continu ed",
+ "ot ion",
+ "oti on",
+ "ca che",
+ "c ache",
+ "dis play",
+ "disp lay",
+ "▁prote ct",
+ "▁prot ect",
+ "am s",
+ "a ms",
+ "▁p ow",
+ "▁po w",
+ "▁ pow",
+ "CT ION",
+ "C TION",
+ "▁M ac",
+ "▁Ma c",
+ "▁ Mac",
+ "m o",
+ "х а",
+ "▁d istance",
+ "▁di stance",
+ "▁dist ance",
+ "▁ distance",
+ "▁T ime",
+ "▁Tim e",
+ "▁Ti me",
+ "▁ Time",
+ "g i",
+ "▁s equ",
+ "▁se qu",
+ "▁seq u",
+ "▁ sequ",
+ "T arget",
+ "с ле",
+ "Ser ver",
+ "Serv er",
+ "▁w ide",
+ "▁wid e",
+ "▁ wide",
+ "cl ose",
+ "clos e",
+ "▁c ru",
+ "▁cr u",
+ "Ex t",
+ "E xt",
+ "▁s elect",
+ "▁se lect",
+ "▁sel ect",
+ "▁sele ct",
+ "▁ select",
+ "▁pat tern",
+ "▁ pattern",
+ "\") );",
+ "\")) ;",
+ "\" ));",
+ "Pro vider",
+ "Prov ider",
+ "UR L",
+ "U RL",
+ "▁g reen",
+ "▁gr een",
+ "▁gre en",
+ "▁ green",
+ "▁wait ing",
+ "▁wa iting",
+ "pro to",
+ "pr oto",
+ "prot o",
+ "▁immedi ately",
+ "▁immediate ly",
+ "com mon",
+ "comm on",
+ "az ione",
+ "azi one",
+ "a zione",
+ "ri ver",
+ "riv er",
+ "rive r",
+ "r iver",
+ "▁s en",
+ "▁se n",
+ "▁ sen",
+ "▁! ==",
+ "▁!= =",
+ "▁Febru ary",
+ "▁Februar y",
+ "ur b",
+ "u rb",
+ "▁S en",
+ "▁Se n",
+ "de st",
+ "des t",
+ "d est",
+ "< ?",
+ "▁ed ge",
+ "▁ edge",
+ "▁m ais",
+ "▁ma is",
+ "▁mai s",
+ "gor ith",
+ "cp u",
+ "c pu",
+ "▁educ ation",
+ "▁associ ated",
+ "▁associate d",
+ "No ne",
+ "Non e",
+ "N one",
+ "h i",
+ "▁p oor",
+ "▁po or",
+ "se m",
+ "s em",
+ "▁W il",
+ "▁Wi l",
+ "▁b ud",
+ "▁bu d",
+ "▁ bud",
+ "▁a uch",
+ "▁au ch",
+ "▁ auch",
+ "el ler",
+ "ell er",
+ "elle r",
+ "▁L ife",
+ "▁Li fe",
+ "▁ Life",
+ "▁f iles",
+ "▁fil es",
+ "▁file s",
+ "▁fi les",
+ "▁ files",
+ "▁le ading",
+ "▁lead ing",
+ "▁ leading",
+ "▁ob tain",
+ "▁obt ain",
+ "▁J ul",
+ "▁Ju l",
+ "at ory",
+ "ator y",
+ "ato ry",
+ "г у",
+ "it able",
+ "ita ble",
+ "i table",
+ "▁on to",
+ "▁ont o",
+ "▁ onto",
+ "▁b orn",
+ "▁bo rn",
+ "▁bor n",
+ "▁ born",
+ "or em",
+ "ore m",
+ "o rem",
+ "▁Stre et",
+ "▁m aint",
+ "▁main t",
+ "▁ma int",
+ "▁mai nt",
+ "Param s",
+ "Par ams",
+ "ri p",
+ "r ip",
+ "▁S T",
+ "▁ ST",
+ "u v",
+ "ma in",
+ "m ain",
+ "▁re cent",
+ "▁rec ent",
+ "▁rece nt",
+ "We b",
+ "W eb",
+ "ov a",
+ "o va",
+ "ц а",
+ "ais e",
+ "ai se",
+ "a ise",
+ "yle s",
+ "yl es",
+ "y les",
+ "▁de scribed",
+ "▁desc ribed",
+ "▁describ ed",
+ "▁describe d",
+ "▁begin ning",
+ "▁D ay",
+ "▁Da y",
+ "▁ Day",
+ "▁V ol",
+ "▁Vo l",
+ "▁ Vol",
+ "▁h uge",
+ "▁hug e",
+ "Ha s",
+ "H as",
+ "an cy",
+ "anc y",
+ "He ader",
+ "Head er",
+ "▁a ren",
+ "▁are n",
+ "▁ar en",
+ "▁ aren",
+ "ва н",
+ "в ан",
+ "▁en sure",
+ "▁ens ure",
+ "▁ ensure",
+ "▁p et",
+ "▁pe t",
+ "▁ pet",
+ "mu lt",
+ "mul t",
+ "m ult",
+ "▁L ike",
+ "▁Li ke",
+ "▁ Like",
+ "▁man agement",
+ "▁manage ment",
+ "▁ management",
+ "P S",
+ "wh ile",
+ "▁back ground",
+ "▁ background",
+ "ount er",
+ "oun ter",
+ "o unter",
+ "bo ol",
+ "b ool",
+ "F C",
+ "N um",
+ "R L",
+ "▁ex cl",
+ "▁exc l",
+ "▁e ye",
+ "▁ey e",
+ "im g",
+ "i mg",
+ "▁r om",
+ "▁ro m",
+ "▁ rom",
+ "▁H el",
+ "▁He l",
+ "▁ Hel",
+ "Opt ion",
+ "O ption",
+ "▁stop ped",
+ "▁sto pped",
+ "▁th read",
+ "▁thr ead",
+ "▁ thread",
+ "to type",
+ "tot ype",
+ "t otype",
+ ")) )",
+ ") ))",
+ "▁st age",
+ "▁stag e",
+ "▁sta ge",
+ "▁ stage",
+ "▁ü ber",
+ "▁ über",
+ "▁al though",
+ "▁ although",
+ "Type s",
+ "Ty pes",
+ "Typ es",
+ "T ypes",
+ "▁O h",
+ "▁ Oh",
+ "▁e ight",
+ "▁ eight",
+ "▁de scription",
+ "▁des cription",
+ "▁ description",
+ "' '",
+ "ö n",
+ "▁sur face",
+ "▁surf ace",
+ "▁ surface",
+ "▁Intern ational",
+ "▁ch arg",
+ "▁char g",
+ "▁cha rg",
+ "▁ charg",
+ "▁col lection",
+ "▁coll ection",
+ "▁collect ion",
+ "▁colle ction",
+ "▁ collection",
+ "▁us ers",
+ "▁use rs",
+ "▁user s",
+ "▁ users",
+ "▁ob vious",
+ "▁cent ury",
+ "▁ century",
+ "ic ks",
+ "ick s",
+ "i cks",
+ "▁art icle",
+ "▁artic le",
+ "▁ article",
+ "▁\" \\",
+ "▁ \"\\",
+ "di m",
+ "d im",
+ "▁s in",
+ "▁si n",
+ "▁ sin",
+ "en ge",
+ "eng e",
+ "Cont rol",
+ "▁com mit",
+ "▁comm it",
+ "▁ commit",
+ "ens ity",
+ "▁t ra",
+ "▁tr a",
+ "▁ tra",
+ "cript or",
+ "▁N OT",
+ "▁NO T",
+ "▁ NOT",
+ "we ll",
+ "w ell",
+ "▁M ichael",
+ "▁Mich ael",
+ "▁n od",
+ "▁no d",
+ "▁ nod",
+ "▁m ort",
+ "▁mor t",
+ "▁mo rt",
+ "iv o",
+ "i vo",
+ "is ation",
+ "▁P o",
+ "▁ Po",
+ "▁P aris",
+ "▁Par is",
+ "▁Pa ris",
+ "▁ad ministr",
+ "▁admin istr",
+ "▁ administr",
+ "bu rg",
+ "bur g",
+ "b urg",
+ "cd ot",
+ "c dot",
+ "▁mil itary",
+ "▁milit ary",
+ "▁militar y",
+ "▁B est",
+ "▁Be st",
+ "▁Bes t",
+ "▁ Best",
+ "▁К а",
+ "▁ Ка",
+ "IN E",
+ "I NE",
+ "▁through out",
+ "S l",
+ "▁im pl",
+ "▁imp l",
+ "▁ impl",
+ "cont rol",
+ "contr ol",
+ "▁ Ч",
+ "▁u it",
+ "▁ui t",
+ "▁ uit",
+ "▁un signed",
+ "▁uns igned",
+ "▁ unsigned",
+ "▁M ary",
+ "▁Mar y",
+ "▁Ma ry",
+ "Ch ar",
+ "C har",
+ "м і",
+ "▁th reat",
+ "▁c ourt",
+ "▁co urt",
+ "▁cour t",
+ "▁cou rt",
+ "▁ court",
+ "vi lle",
+ "vil le",
+ "v ille",
+ "▁ ш",
+ "▁C am",
+ "▁Ca m",
+ "▁ Cam",
+ ". \r",
+ "▁current ly",
+ "▁curr ently",
+ "ro t",
+ "r ot",
+ "▁D ate",
+ "▁Da te",
+ "▁Dat e",
+ "▁ Date",
+ "▁s hit",
+ "▁sh it",
+ "▁ shit",
+ "▁$ {\\",
+ "▁${ \\",
+ "un n",
+ "u nn",
+ "U s",
+ "▁b uffer",
+ "▁buff er",
+ "▁buf fer",
+ "▁ buffer",
+ "▁s ont",
+ "▁so nt",
+ "▁son t",
+ "▁let ter",
+ "▁lett er",
+ "▁ letter",
+ "in ated",
+ "ina ted",
+ "inate d",
+ "Ch ange",
+ "▁h ref",
+ "▁hr ef",
+ "▁ href",
+ "▁l ack",
+ "▁la ck",
+ "▁lac k",
+ "▁o il",
+ "▁C ons",
+ "▁Con s",
+ "▁Co ns",
+ "▁ Cons",
+ "▁J er",
+ "▁Je r",
+ "BU G",
+ "B UG",
+ "if orn",
+ "▁pro perties",
+ "▁proper ties",
+ "▁ properties",
+ "▁r andom",
+ "▁ran dom",
+ "▁rand om",
+ "▁ random",
+ "▁br other",
+ "▁bro ther",
+ "▁p iece",
+ "▁pie ce",
+ "▁ piece",
+ "б у",
+ "ist ics",
+ "istic s",
+ "isti cs",
+ "▁techn ology",
+ "gl obal",
+ "glob al",
+ "▁trans form",
+ "▁ transform",
+ "er d",
+ "e rd",
+ "▁B ecause",
+ "▁ Because",
+ "PE CT",
+ "P ECT",
+ "pr et",
+ "pre t",
+ "p ret",
+ "▁го ду",
+ "▁год у",
+ "▁M et",
+ "▁Me t",
+ "▁ Met",
+ "▁p sy",
+ "▁ps y",
+ "▁ psy",
+ "▁о д",
+ "▁g od",
+ "▁go d",
+ "▁ god",
+ "▁D el",
+ "▁De l",
+ "▁ Del",
+ "base d",
+ "ba sed",
+ "bas ed",
+ "b ased",
+ "▁v oor",
+ "▁vo or",
+ "▁C all",
+ "▁Cal l",
+ "▁Ca ll",
+ "▁ Call",
+ "S A",
+ "▁fil ter",
+ "▁ filter",
+ "▁incl udes",
+ "▁includ es",
+ "▁include s",
+ "▁inclu des",
+ "▁ includes",
+ "olut ions",
+ "olution s",
+ "f d",
+ "▁w ind",
+ "▁win d",
+ "▁ wind",
+ "▁б о",
+ "▁ бо",
+ "▁ab ility",
+ "▁ ability",
+ "ca rd",
+ "car d",
+ "c ard",
+ "▁n umer",
+ "▁num er",
+ "▁nu mer",
+ "▁ numer",
+ "add ress",
+ "addr ess",
+ "▁go al",
+ "ash ington",
+ "ashing ton",
+ "▁s light",
+ "▁sl ight",
+ "ab a",
+ "a ba",
+ "▁L og",
+ "▁Lo g",
+ "▁ Log",
+ "Set tings",
+ "Setting s",
+ "ad ow",
+ "ado w",
+ "▁p i",
+ "▁ pi",
+ "ir ing",
+ "iri ng",
+ "i ring",
+ "F T",
+ "▁number s",
+ "▁num bers",
+ "con f",
+ "co nf",
+ "ta sk",
+ "t ask",
+ "▁î n",
+ "т ы",
+ "▁re ceive",
+ "▁rece ive",
+ "▁r oot",
+ "▁ro ot",
+ "▁ root",
+ "▁Ind ia",
+ "pat ch",
+ "p atch",
+ "é l",
+ "▁sum mer",
+ "▁method s",
+ "▁ methods",
+ "▁pl aces",
+ "▁place s",
+ "▁plac es",
+ "▁М а",
+ "▁ Ма",
+ "▁cap ital",
+ "▁capit al",
+ "▁ev idence",
+ "▁G erman",
+ "▁Germ an",
+ "▁Ger man",
+ "\\ ,",
+ "D A",
+ "ec ute",
+ "ecut e",
+ "col umn",
+ "▁fun ctions",
+ "▁function s",
+ "▁ functions",
+ "▁c ounter",
+ "▁co unter",
+ "▁coun ter",
+ "▁count er",
+ "▁ counter",
+ "▁ar ms",
+ "▁arm s",
+ "▁ arms",
+ "▁f eed",
+ "▁fe ed",
+ "▁fee d",
+ "▁ feed",
+ "ve y",
+ "v ey",
+ "he nt",
+ "hen t",
+ "h ent",
+ "MA X",
+ "M AX",
+ "▁ac qu",
+ "▁app ly",
+ "▁ap ply",
+ "▁appl y",
+ "▁ apply",
+ "▁hus band",
+ "▁k illed",
+ "▁kill ed",
+ "▁kil led",
+ "▁S pec",
+ "▁Sp ec",
+ "▁Spe c",
+ "▁ Spec",
+ "ent ity",
+ "enti ty",
+ "▁e arlier",
+ "▁M iss",
+ "▁Mi ss",
+ "▁Mis s",
+ "▁ Miss",
+ "▁set ting",
+ "▁sett ing",
+ "▁ setting",
+ "it ect",
+ "ite ct",
+ "▁d ed",
+ "▁de d",
+ "▁ ded",
+ "Ro w",
+ "R ow",
+ "▁r an",
+ "▁ra n",
+ "▁ ran",
+ "▁Y es",
+ "▁Ye s",
+ "▁ Yes",
+ "▁fin ancial",
+ "▁financ ial",
+ "s ession",
+ "le ar",
+ "l ear",
+ "is hing",
+ "ish ing",
+ "ishi ng",
+ "▁ne arly",
+ "▁near ly",
+ "▁d ur",
+ "▁du r",
+ "▁m achine",
+ "▁mach ine",
+ "▁ machine",
+ "xf f",
+ "x ff",
+ "br o",
+ "b ro",
+ "▁s ymbol",
+ "▁sym bol",
+ "▁ symbol",
+ "land s",
+ "lan ds",
+ "l ands",
+ "Ac c",
+ "A cc",
+ "d i",
+ "▁Rober t",
+ "▁Ro bert",
+ "▁Rob ert",
+ "pro p",
+ "pr op",
+ "p rop",
+ "ur ity",
+ "uri ty",
+ "▁# ####",
+ "▁## ###",
+ "▁### ##",
+ "▁#### #",
+ "▁walk ed",
+ "▁wal ked",
+ "▁intern ational",
+ "▁internation al",
+ "▁ Е",
+ "Y es",
+ "▁re lease",
+ "▁rele ase",
+ "▁ release",
+ "▁start ing",
+ "▁star ting",
+ "st atic",
+ "stat ic",
+ "▁b ei",
+ "▁be i",
+ "al low",
+ "all ow",
+ "allo w",
+ "▁Pe ople",
+ "▁ People",
+ "e z",
+ "▁param eter",
+ "▁ parameter",
+ "C ache",
+ "▁$ $",
+ "▁ $$",
+ "amp ions",
+ "ampion s",
+ "▁M er",
+ "▁Me r",
+ "▁ Mer",
+ "▁k om",
+ "▁ko m",
+ "▁ kom",
+ "le ted",
+ "let ed",
+ "lete d",
+ "l eted",
+ "oi s",
+ "o is",
+ "▁O pen",
+ "▁Op en",
+ "▁ Open",
+ "ty pes",
+ "type s",
+ "typ es",
+ "t ypes",
+ "▁f ue",
+ "▁fu e",
+ "ac ters",
+ "act ers",
+ "acter s",
+ "▁re ference",
+ "▁refer ence",
+ "▁ reference",
+ "Equ als",
+ "Equal s",
+ "Eq uals",
+ "▁a ware",
+ "▁aw are",
+ "▁ aware",
+ "▁h ol",
+ "▁ho l",
+ "▁ hol",
+ "▁de mand",
+ "▁dem and",
+ "lo r",
+ "l or",
+ "▁v eh",
+ "▁ve h",
+ "▁ veh",
+ "▁not ice",
+ "▁ notice",
+ "▁com ponent",
+ "▁compon ent",
+ "▁ component",
+ "f n",
+ "▁anal ysis",
+ "▁analy sis",
+ "▁analys is",
+ "▁ analysis",
+ "mat ch",
+ "m atch",
+ "▁effect ive",
+ "▁ effective",
+ "pro duct",
+ "produ ct",
+ "prod uct",
+ "ни к",
+ "▁le gal",
+ "▁leg al",
+ "▁ legal",
+ "е й",
+ "se mb",
+ "sem b",
+ "s emb",
+ "▁loc ated",
+ "▁locate d",
+ "▁с у",
+ "▁ су",
+ "Q L",
+ "in ct",
+ "inc t",
+ "et o",
+ "e to",
+ "Dr aw",
+ "D raw",
+ "▁sc ale",
+ "▁scal e",
+ "▁ scale",
+ "ро в",
+ "р ов",
+ "▁w ants",
+ "▁want s",
+ "H ow",
+ "▁w el",
+ "▁we l",
+ "is ions",
+ "ision s",
+ "isi ons",
+ "▁de liver",
+ "▁del iver",
+ "un der",
+ "und er",
+ "unde r",
+ "u nder",
+ "▁d eb",
+ "▁de b",
+ "▁j u",
+ "▁ ju",
+ "val ues",
+ "value s",
+ "▁s ister",
+ "▁si ster",
+ "▁sist er",
+ "ко в",
+ "к ов",
+ "▁C reate",
+ "▁Creat e",
+ "▁Cre ate",
+ "▁ Create",
+ "▁I nc",
+ "▁In c",
+ "▁a ux",
+ "▁au x",
+ "▁ aux",
+ "▁Wh ite",
+ "▁Whit e",
+ "▁ White",
+ "Me nu",
+ "Men u",
+ "M enu",
+ "au d",
+ "a ud",
+ "re source",
+ "res ource",
+ "▁c ab",
+ "▁ca b",
+ "▁l if",
+ "▁li f",
+ "▁ lif",
+ "▁c ulture",
+ "▁cult ure",
+ "ic he",
+ "ich e",
+ "i che",
+ "▁wh atever",
+ "▁what ever",
+ "▁de signed",
+ "▁des igned",
+ "▁design ed",
+ "▁re pe",
+ "▁rep e",
+ "▁M ont",
+ "▁Mon t",
+ "▁Mo nt",
+ "▁ Mont",
+ "▁ch arge",
+ "▁char ge",
+ "▁charg e",
+ "▁ charge",
+ "Name s",
+ "Na mes",
+ "N ames",
+ "▁in sp",
+ "▁ins p",
+ "▁custom ers",
+ "▁customer s",
+ "os a",
+ "o sa",
+ "▁d aughter",
+ "▁E ast",
+ "E Q",
+ "▁o pin",
+ "▁op in",
+ "▁F re",
+ "▁Fr e",
+ "▁se ek",
+ "▁see k",
+ "▁ seek",
+ "▁p ush",
+ "▁pu sh",
+ "▁ push",
+ "▁n av",
+ "▁na v",
+ "▁ nav",
+ "▁b urn",
+ "▁bu rn",
+ "▁bur n",
+ "▁ burn",
+ "ar den",
+ "ard en",
+ "arde n",
+ "ha sh",
+ "has h",
+ "h ash",
+ "▁opportun ity",
+ "▁M at",
+ "▁Ma t",
+ "▁ Mat",
+ "oy al",
+ "oya l",
+ "o yal",
+ "▁p un",
+ "▁pu n",
+ "sc ale",
+ "scal e",
+ "yn amic",
+ "ynam ic",
+ "yna mic",
+ "▁T ype",
+ "▁Ty pe",
+ "▁Typ e",
+ "▁ Type",
+ "il ing",
+ "ili ng",
+ "i ling",
+ "▁qu ery",
+ "▁que ry",
+ "▁quer y",
+ "▁ query",
+ "▁m ist",
+ "▁mis t",
+ "▁mi st",
+ "ro r",
+ "r or",
+ "for ce",
+ "▁On ce",
+ "▁ Once",
+ "▁med ical",
+ "▁medic al",
+ "▁medi cal",
+ "li e",
+ "l ie",
+ "▁stud ent",
+ "▁ student",
+ "ed eral",
+ "eder al",
+ "ede ral",
+ "▁l ov",
+ "▁lo v",
+ "▁ lov",
+ "if orm",
+ "i form",
+ "▁al tern",
+ "▁alt ern",
+ "▁alter n",
+ "▁ altern",
+ "bi n",
+ "b in",
+ "od er",
+ "ode r",
+ "o der",
+ "▁return s",
+ "▁ returns",
+ "reg ister",
+ "ut s",
+ "u ts",
+ "C I",
+ "▁T or",
+ "▁To r",
+ "▁ Tor",
+ "C R",
+ "▁L os",
+ "▁Lo s",
+ "▁ Los",
+ "am ily",
+ "ami ly",
+ "amil y",
+ "air e",
+ "ai re",
+ "a ire",
+ "++ ;",
+ "Cont roller",
+ "Control ler",
+ "wi de",
+ "wid e",
+ "w ide",
+ "x x",
+ "row ser",
+ "rows er",
+ "▁B ook",
+ "▁Bo ok",
+ "▁ Book",
+ "Cont ainer",
+ "pl oad",
+ "plo ad",
+ "p load",
+ "▁E v",
+ "▁ Ev",
+ "▁t al",
+ "▁ta l",
+ "▁ tal",
+ "▁the ory",
+ "eqn array",
+ "б е",
+ "▁rep orted",
+ "▁report ed",
+ "▁me aning",
+ "▁mean ing",
+ "▁s y",
+ "▁ sy",
+ "ri be",
+ "rib e",
+ "r ibe",
+ "ic ate",
+ "ica te",
+ "ho ld",
+ "hol d",
+ "h old",
+ "▁of fers",
+ "▁off ers",
+ "▁offer s",
+ "▁t empl",
+ "▁tem pl",
+ "▁temp l",
+ "cs s",
+ "c ss",
+ "▁p icture",
+ "▁pict ure",
+ "▁ picture",
+ "▁a sync",
+ "▁as ync",
+ "▁ async",
+ "▁st ock",
+ "▁sto ck",
+ "▁ stock",
+ "▁in ternal",
+ "▁inter nal",
+ "▁intern al",
+ "▁ internal",
+ "t i",
+ "B O",
+ "V er",
+ "с по",
+ "▁d emon",
+ "▁de mon",
+ "▁dem on",
+ "▁demo n",
+ "▁l augh",
+ "▁la ugh",
+ "▁laug h",
+ "▁E nd",
+ "▁En d",
+ "▁ End",
+ "▁k on",
+ "▁ko n",
+ "▁ kon",
+ "▁ide as",
+ "▁idea s",
+ "▁c andid",
+ "▁can did",
+ "▁cand id",
+ "Me m",
+ "M em",
+ "iz z",
+ "i zz",
+ "re fix",
+ "ref ix",
+ "▁A ND",
+ "▁AN D",
+ "▁ AND",
+ "eg en",
+ "e gen",
+ "E l",
+ "▁camp aign",
+ "H ttp",
+ "▁R ob",
+ "▁Ro b",
+ "▁ Rob",
+ "д і",
+ "▁b ul",
+ "▁bu l",
+ "▁ bul",
+ "▁К о",
+ "▁ Ко",
+ "▁count ries",
+ "▁countr ies",
+ "» .",
+ "▁ex pression",
+ "▁exp ression",
+ "▁express ion",
+ "▁expr ession",
+ "▁ expression",
+ "▁Eng land",
+ "s f",
+ "▁certain ly",
+ "ag en",
+ "age n",
+ "a gen",
+ "▁ч а",
+ "▁ ча",
+ "▁A NY",
+ "▁AN Y",
+ "▁ ANY",
+ "▁conne ct",
+ "▁conn ect",
+ "▁ connect",
+ "F E",
+ "▁and roid",
+ "▁ android",
+ "▁G old",
+ "▁Go ld",
+ "▁Gol d",
+ "▁ Gold",
+ "▁op pos",
+ "▁opp os",
+ "ov ern",
+ "ove rn",
+ "over n",
+ "o vern",
+ "▁Com mun",
+ "▁Comm un",
+ ", _",
+ "as ion",
+ "asi on",
+ "L a",
+ "▁f irm",
+ "▁fi rm",
+ "▁fir m",
+ "▁Al though",
+ "▁G ood",
+ "▁Go od",
+ "▁ Good",
+ "▁L aw",
+ "▁La w",
+ "er ve",
+ "erv e",
+ "▁b rand",
+ "▁br and",
+ "▁bra nd",
+ "▁ brand",
+ "M in",
+ "fil l",
+ "fi ll",
+ "f ill",
+ "'] ,",
+ "' ],",
+ "▁J ew",
+ "▁Je w",
+ "il er",
+ "ile r",
+ "i ler",
+ "in gle",
+ "ing le",
+ "it hub",
+ "ith ub",
+ "▁D iv",
+ "▁Di v",
+ "▁ Div",
+ "▁c ert",
+ "▁ce rt",
+ "▁cer t",
+ "▁ cert",
+ "He ight",
+ "H eight",
+ "ra el",
+ "r ael",
+ "The re",
+ "Th ere",
+ "T here",
+ "it ute",
+ "itut e",
+ "itu te",
+ "▁a maz",
+ "▁am az",
+ "▁ amaz",
+ "lo ok",
+ "l ook",
+ "▁S E",
+ "▁ SE",
+ "▁j o",
+ "▁ jo",
+ "▁pull ed",
+ "▁pul led",
+ "▁re sources",
+ "▁res ources",
+ "▁resource s",
+ "▁ resources",
+ "▁M ax",
+ "▁Ma x",
+ "▁ Max",
+ "▁ag reed",
+ "▁agree d",
+ "▁agre ed",
+ "as y",
+ "a sy",
+ "▁treat ment",
+ "\"> ",
+ "\">< /",
+ "\" >",
+ "ма н",
+ "м ан",
+ "▁E rr",
+ "▁Er r",
+ "▁ Err",
+ "or ig",
+ "ori g",
+ "o rig",
+ "co s",
+ "c os",
+ "▁May be",
+ "▁ Maybe",
+ "ot al",
+ "ota l",
+ "o tal",
+ "▁tr ain",
+ "▁tra in",
+ "▁ train",
+ "▁S ervice",
+ "▁Serv ice",
+ "▁ Service",
+ "▁i h",
+ "▁ ih",
+ "▁sp irit",
+ "▁spir it",
+ "Com p",
+ "Co mp",
+ "C omp",
+ "sq rt",
+ "▁b road",
+ "▁br oad",
+ "▁bro ad",
+ "▁ broad",
+ "} [",
+ "▁sh ape",
+ "▁sha pe",
+ "▁ shape",
+ "▁d oc",
+ "▁do c",
+ "▁ doc",
+ "ho w",
+ "h ow",
+ "▁t ag",
+ "▁ta g",
+ "▁ tag",
+ "ata log",
+ "atal og",
+ "s d",
+ "▁me as",
+ "▁Р о",
+ "▁ex ception",
+ "▁except ion",
+ "▁ exception",
+ "▁T w",
+ "▁ Tw",
+ "▁interest ing",
+ "AT A",
+ "A TA",
+ "▁R el",
+ "▁Re l",
+ "▁ Rel",
+ "á r",
+ "▁use ful",
+ "use um",
+ "▁b ottom",
+ "▁bott om",
+ "▁bot tom",
+ "▁ bottom",
+ "▁other wise",
+ "▁ag ree",
+ "▁agre e",
+ "ch t",
+ "c ht",
+ "th en",
+ "the n",
+ "t hen",
+ "▁signific ant",
+ "} /",
+ "▁ch annel",
+ "▁ channel",
+ "ic ial",
+ "ici al",
+ "icia l",
+ "i cial",
+ "ти в",
+ "var e",
+ "va re",
+ "v are",
+ "▁en ter",
+ "▁ent er",
+ "▁ enter",
+ "En g",
+ "E ng",
+ "u j",
+ "UR E",
+ "U RE",
+ "que ue",
+ "on o",
+ "o no",
+ "▁cont ains",
+ "▁contain s",
+ "▁ contains",
+ "M I",
+ "▁n ation",
+ "▁nat ion",
+ "▁r ules",
+ "▁rule s",
+ "▁ru les",
+ "▁rul es",
+ "▁ rules",
+ "fo l",
+ "f ol",
+ "▁p a",
+ "▁ pa",
+ "ar p",
+ "a rp",
+ "▁qu iet",
+ "▁qui et",
+ "▁t hus",
+ "▁th us",
+ "ip ped",
+ "ipp ed",
+ "i pped",
+ "an not",
+ "ann ot",
+ "anno t",
+ "ud es",
+ "ude s",
+ "u des",
+ "() :",
+ "( ):",
+ "name s",
+ "na mes",
+ "nam es",
+ "n ames",
+ "▁com pos",
+ "▁comp os",
+ "▁in j",
+ "un a",
+ "u na",
+ "bin d",
+ "bi nd",
+ "b ind",
+ "▁f ully",
+ "▁full y",
+ "▁ful ly",
+ "▁ fully",
+ "ra s",
+ "r as",
+ "Util s",
+ "Ut ils",
+ "an ges",
+ "ang es",
+ "ange s",
+ "du le",
+ "d ule",
+ "▁Christ ian",
+ "▁re ve",
+ "▁r eve",
+ "▁rev e",
+ "än d",
+ "ä nd",
+ "▁col lect",
+ "▁coll ect",
+ "▁colle ct",
+ "▁ collect",
+ "▁cele br",
+ "an da",
+ "and a",
+ "í n",
+ "jo in",
+ "j oin",
+ "▁p aid",
+ "▁pa id",
+ "▁ paid",
+ "Co re",
+ "Cor e",
+ "C ore",
+ "G e",
+ ". $",
+ "▁f if",
+ "▁fi f",
+ "▁ fif",
+ "▁u ma",
+ "▁um a",
+ "▁ uma",
+ "▁ ~",
+ "erv ices",
+ "ervice s",
+ "▁rec ently",
+ "▁recent ly",
+ "de sc",
+ "des c",
+ "d esc",
+ "▁he avy",
+ "▁heav y",
+ "▁r ule",
+ "▁ru le",
+ "▁rul e",
+ "▁ rule",
+ "▁P lease",
+ "▁Ple ase",
+ "▁ Please",
+ "ps i",
+ "p si",
+ "▁con sole",
+ "▁cons ole",
+ "▁ console",
+ "▁f ort",
+ "▁for t",
+ "▁fo rt",
+ "▁ fort",
+ ". \\",
+ "▁W ashington",
+ "▁g ar",
+ "▁ga r",
+ "▁ gar",
+ "▁G roup",
+ "▁Gr oup",
+ "▁Gro up",
+ "▁ Group",
+ "▁inter view",
+ "an ned",
+ "ann ed",
+ "anne d",
+ "sq l",
+ "s ql",
+ "▁a nc",
+ "▁an c",
+ "▁ anc",
+ "ј а",
+ "P ack",
+ "▁Cl ub",
+ "▁m ask",
+ "▁ma sk",
+ "▁mas k",
+ "▁ mask",
+ "▁con cept",
+ "▁conce pt",
+ "▁[ '",
+ "▁ ['",
+ "▁se lected",
+ "▁select ed",
+ "▁sele cted",
+ "▁ selected",
+ "▁U se",
+ "▁Us e",
+ "▁ Use",
+ "▁e le",
+ "▁el e",
+ "▁ ele",
+ "ear s",
+ "ea rs",
+ "e ars",
+ "▁r ace",
+ "▁rac e",
+ "▁ra ce",
+ "h y",
+ "O m",
+ "▁st eps",
+ "▁ste ps",
+ "▁step s",
+ "▁ steps",
+ "il a",
+ "i la",
+ "es ts",
+ "est s",
+ "e sts",
+ "ed s",
+ "e ds",
+ "▁stre et",
+ "ne rs",
+ "ner s",
+ "n ers",
+ "▁b irth",
+ "po p",
+ "p op",
+ "▁ ли",
+ "M B",
+ "к ра",
+ "ci r",
+ "c ir",
+ "eps ilon",
+ "e psilon",
+ "▁con stant",
+ "▁const ant",
+ "▁ constant",
+ "qu es",
+ "que s",
+ "q ues",
+ "ad as",
+ "ada s",
+ "a das",
+ "▁kn ows",
+ "▁know s",
+ "▁P y",
+ "▁ Py",
+ "cl es",
+ "cle s",
+ "c les",
+ "▁c it",
+ "▁ci t",
+ "▁ cit",
+ "▁p air",
+ "▁pa ir",
+ "▁ pair",
+ "in ese",
+ "ine se",
+ "ines e",
+ "▁P eter",
+ "▁Pe ter",
+ "▁Pet er",
+ "▁Pete r",
+ "▁fin ished",
+ "▁finish ed",
+ "▁ finished",
+ "▁m aster",
+ "▁ma ster",
+ "▁mas ter",
+ "▁mast er",
+ "▁ master",
+ "▁tw enty",
+ "▁f ell",
+ "▁fe ll",
+ "▁fel l",
+ "▁cent ral",
+ "▁m es",
+ "▁me s",
+ "▁ mes",
+ "re v",
+ "r ev",
+ "ST AT",
+ "st at",
+ "sta t",
+ "s tat",
+ "▁all ows",
+ "▁allow s",
+ "▁g ro",
+ "▁gr o",
+ "▁ gro",
+ "Cl ick",
+ "C lick",
+ "▁st ories",
+ "▁stor ies",
+ "▁sto ries",
+ "F e",
+ "å r",
+ "▁b aby",
+ "▁bab y",
+ "▁ba by",
+ "en cia",
+ "enc ia",
+ "enci a",
+ "e ncia",
+ "▁e iner",
+ "▁ein er",
+ "▁eine r",
+ "Ar e",
+ "A re",
+ "eb ug",
+ "e bug",
+ "st ore",
+ "sto re",
+ "\", \"",
+ "\" ,\"",
+ "la m",
+ "l am",
+ "▁s v",
+ "▁ sv",
+ "ци и",
+ "NU LL",
+ "N ULL",
+ "▁L eg",
+ "▁Le g",
+ "▁ Leg",
+ "▁m ovie",
+ "▁mov ie",
+ "▁h ous",
+ "▁ho us",
+ "▁learn ed",
+ "▁lear ned",
+ "bo n",
+ "b on",
+ "▁trans fer",
+ "▁ transfer",
+ "iforn ia",
+ "ps ilon",
+ "psi lon",
+ "▁S oft",
+ "▁So ft",
+ "▁Sof t",
+ "▁ Soft",
+ "▁com mer",
+ "▁comm er",
+ "▁comme r",
+ "▁had n",
+ "▁ha dn",
+ "▁E in",
+ "▁T wo",
+ "▁Tw o",
+ "▁ Two",
+ "cr aft",
+ "c raft",
+ "Pro cess",
+ "Proc ess",
+ "▁по д",
+ "ar gin",
+ "arg in",
+ "▁est im",
+ "▁es tim",
+ "▁M em",
+ "▁Me m",
+ "▁ Mem",
+ "ik a",
+ "i ka",
+ "▁T od",
+ "▁To d",
+ "du c",
+ "d uc",
+ "▁d anger",
+ "▁dan ger",
+ "ri ve",
+ "riv e",
+ "r ive",
+ "Do n",
+ "D on",
+ "▁Q ue",
+ "▁Qu e",
+ "▁ Que",
+ "ha l",
+ "h al",
+ "▁m m",
+ "▁ mm",
+ "▁S ur",
+ "▁Su r",
+ "▁ Sur",
+ "Or der",
+ "Ord er",
+ "▁d istribution",
+ "▁distribut ion",
+ "f a",
+ "▁M any",
+ "▁Man y",
+ "▁Ma ny",
+ "▁ Many",
+ "pl icit",
+ "plic it",
+ "Em pty",
+ "Emp ty",
+ "Hand le",
+ "▁t oken",
+ "▁to ken",
+ "▁tok en",
+ "▁ token",
+ "▁e pis",
+ "▁ep is",
+ "▁ass ist",
+ "▁pur pose",
+ "▁ ц",
+ "N U",
+ "id ers",
+ "ide rs",
+ "ider s",
+ "i ders",
+ "ra te",
+ "rat e",
+ "r ate",
+ "The y",
+ "Th ey",
+ "Param eter",
+ "De c",
+ "D ec",
+ "▁str ugg",
+ "▁stru gg",
+ "▁sh oot",
+ "I V",
+ "▁G reat",
+ "▁Gre at",
+ "▁ Great",
+ "▁S il",
+ "▁Si l",
+ "▁ Sil",
+ "▁l oved",
+ "▁lo ved",
+ "▁love d",
+ "▁lov ed",
+ "▁c lick",
+ "▁cl ick",
+ "▁ click",
+ "▁re serv",
+ "▁res erv",
+ "▁в е",
+ "▁ ве",
+ "▁s pread",
+ "▁sp read",
+ "▁spr ead",
+ "▁o g",
+ "▁ og",
+ "▁$ {",
+ "▁ ${",
+ "▁m iles",
+ "▁mil es",
+ "▁mi les",
+ "▁mile s",
+ "▁success ful",
+ "▁ successful",
+ "o j",
+ "▁D irect",
+ "▁Di rect",
+ "▁Dire ct",
+ "▁Dir ect",
+ "▁ Direct",
+ "▁a x",
+ "▁ ax",
+ "▁grow th",
+ "W ork",
+ "▁ch urch",
+ "In st",
+ "Ins t",
+ "IC E",
+ "I CE",
+ "st en",
+ "ste n",
+ "s ten",
+ "ро д",
+ "▁C enter",
+ "▁Cent er",
+ "▁ Center",
+ "se s",
+ "s es",
+ "go t",
+ "g ot",
+ "de lete",
+ "del ete",
+ "▁M a",
+ "▁ Ma",
+ "% %",
+ "▁c row",
+ "▁cr ow",
+ "▁cro w",
+ "D F",
+ "fr ont",
+ "▁b log",
+ "▁bl og",
+ "▁blo g",
+ "▁ blog",
+ "▁comp uter",
+ "▁comput er",
+ "▁compute r",
+ "на я",
+ "▁m ir",
+ "▁mi r",
+ "▁ mir",
+ "▁S uper",
+ "▁Su per",
+ "▁Sup er",
+ "▁ Super",
+ "', '",
+ "' ,'",
+ "▁mult i",
+ "▁mul ti",
+ "▁ multi",
+ "▁g ru",
+ "▁gr u",
+ "▁ gru",
+ "▁J o",
+ "▁ Jo",
+ "▁Can ada",
+ "▁Canad a",
+ "▁Th omas",
+ "▁Thom as",
+ "▁large r",
+ "▁larg er",
+ "▁com par",
+ "▁comp ar",
+ "▁ compar",
+ "Cur rent",
+ "th at",
+ "tha t",
+ "t hat",
+ "▁d rop",
+ "▁dr op",
+ "▁dro p",
+ "▁ drop",
+ "ен т",
+ "▁Re public",
+ "▁Rep ublic",
+ "▁Repub lic",
+ "▁d ise",
+ "▁dis e",
+ "▁di se",
+ "▁effect s",
+ "▁girl s",
+ "▁gir ls",
+ "en cies",
+ "enc ies",
+ "enci es",
+ "el lig",
+ "ell ig",
+ "elli g",
+ "▁N ote",
+ "▁No te",
+ "▁Not e",
+ "▁ Note",
+ "▁Ass oci",
+ "▁ Associ",
+ "▁u ses",
+ "▁us es",
+ "▁use s",
+ "▁ uses",
+ "el led",
+ "ell ed",
+ "elle d",
+ "▁w arm",
+ "▁war m",
+ "▁wa rm",
+ "th read",
+ "fo nt",
+ "fon t",
+ "f ont",
+ "▁z um",
+ "▁zu m",
+ "▁follow s",
+ "▁w hom",
+ "▁wh om",
+ "▁who m",
+ "T A",
+ "▁w ild",
+ "▁A R",
+ "▁ AR",
+ "ia ble",
+ "i able",
+ "▁Tr ue",
+ "▁Tru e",
+ "▁ True",
+ "Pos ition",
+ "▁s ell",
+ "▁se ll",
+ "▁sel l",
+ "ch er",
+ "che r",
+ "c her",
+ "▁B us",
+ "▁Bu s",
+ "▁ Bus",
+ "▁le an",
+ "▁ lean",
+ "AC E",
+ "A CE",
+ "▁s erved",
+ "▁ser ved",
+ "▁serv ed",
+ "▁serve d",
+ "h w",
+ "▁C ur",
+ "▁Cu r",
+ "▁ Cur",
+ "▁n orth",
+ "▁nor th",
+ "▁nort h",
+ "Da t",
+ "D at",
+ "▁> >",
+ "▁ >>",
+ "com mand",
+ "comm and",
+ "at z",
+ "a tz",
+ "▁m al",
+ "▁ma l",
+ "▁ mal",
+ "ста в",
+ "▁P ress",
+ "▁Pr ess",
+ "▁Pres s",
+ "▁Pre ss",
+ "▁ Press",
+ "▁char acters",
+ "▁character s",
+ "▁z ero",
+ "▁ze ro",
+ "▁ zero",
+ "AG E",
+ "A GE",
+ "rap per",
+ "▁kit chen",
+ "am ing",
+ "ami ng",
+ "amin g",
+ "a ming",
+ "▁re str",
+ "▁r estr",
+ "▁res tr",
+ "▁rest r",
+ "X X",
+ "▁Col lege",
+ "▁Ar ray",
+ "▁Arr ay",
+ "▁ Array",
+ "▁f resh",
+ "▁fr esh",
+ "▁fre sh",
+ "▁fres h",
+ "▁sh ift",
+ "▁ shift",
+ "▁spec ified",
+ "pl ete",
+ "ple te",
+ "plet e",
+ "p lete",
+ "IT E",
+ "I TE",
+ "▁C amp",
+ "▁Cam p",
+ "▁Ca mp",
+ "▁ Camp",
+ "ri al",
+ "ria l",
+ "r ial",
+ "c b",
+ "▁T H",
+ "▁ TH",
+ "I B",
+ "os en",
+ "ose n",
+ "o sen",
+ "▁ ú",
+ "▁par ams",
+ "▁param s",
+ "▁para ms",
+ "▁ params",
+ "ign ment",
+ "ad ding",
+ "add ing",
+ "▁deg ree",
+ "▁ degree",
+ "Loc al",
+ "Lo cal",
+ "L ocal",
+ "O h",
+ "▁z ur",
+ "▁zu r",
+ "▁level s",
+ "▁lev els",
+ "C S",
+ "fin ished",
+ "finish ed",
+ "C ase",
+ "ri age",
+ "ria ge",
+ "Vec tor",
+ "V ector",
+ "▁s ea",
+ "▁se a",
+ "▁ sea",
+ "ant ic",
+ "anti c",
+ "▁Le ague",
+ "▁there fore",
+ "▁ther efore",
+ "On e",
+ "O ne",
+ "Re turn",
+ "Ret urn",
+ "R eturn",
+ "Acc ess",
+ "Ac cess",
+ "A ccess",
+ "va s",
+ "v as",
+ "▁о с",
+ "▁r at",
+ "▁ra t",
+ "▁ rat",
+ "Bi g",
+ "B ig",
+ "▁be havior",
+ "▁behav ior",
+ "▁behavi or",
+ "k r",
+ "▁un defined",
+ "▁und efined",
+ "▁ undefined",
+ "▁E s",
+ "▁ Es",
+ "▁appe ared",
+ "▁appear ed",
+ "el es",
+ "ele s",
+ "e les",
+ "▁W AR",
+ "▁WA R",
+ "▁ WAR",
+ "St at",
+ "S tat",
+ "▁Go ogle",
+ "▁ Google",
+ "▁c redit",
+ "▁cre dit",
+ "▁cr edit",
+ "▁cred it",
+ "▁F ile",
+ "▁Fil e",
+ "▁Fi le",
+ "▁ File",
+ "an ging",
+ "ang ing",
+ "ho use",
+ "hou se",
+ "h ouse",
+ "rom ise",
+ "ge nt",
+ "gen t",
+ "g ent",
+ "▁hab it",
+ "▁ha bit",
+ "▁soc iety",
+ "▁soci ety",
+ "▁societ y",
+ "▁enc our",
+ "▁p aint",
+ "▁pain t",
+ "▁pa int",
+ "pe t",
+ "p et",
+ "▁U K",
+ "▁ UK",
+ "aw s",
+ "a ws",
+ "on om",
+ "ono m",
+ "o nom",
+ "G l",
+ "}_ {\\",
+ "}_{ \\",
+ "} _{\\",
+ "el ess",
+ "ele ss",
+ "eles s",
+ "e less",
+ "em y",
+ "e my",
+ "▁C ong",
+ "▁Con g",
+ "▁Co ng",
+ "▁develop ed",
+ "▁im ages",
+ "▁image s",
+ "▁imag es",
+ "▁ images",
+ "▁ ö",
+ "▁f ont",
+ "▁fo nt",
+ "▁fon t",
+ "▁ font",
+ "cl ear",
+ "cle ar",
+ "c lear",
+ "gi n",
+ "g in",
+ "▁L ord",
+ "▁Lo rd",
+ "▁Lor d",
+ "▁trans port",
+ "▁ transport",
+ "▁: :",
+ "▁ ::",
+ "▁c up",
+ "▁cu p",
+ "▁ cup",
+ "ul ate",
+ "ula te",
+ "u late",
+ "▁D uring",
+ "▁Du ring",
+ "▁Dur ing",
+ "pr iv",
+ "p riv",
+ "▁ext rem",
+ "▁extr em",
+ "▁D i",
+ "▁ Di",
+ "▁d oubt",
+ "▁dou bt",
+ "▁doub t",
+ "P y",
+ "if ying",
+ "ify ing",
+ "sp lit",
+ "spl it",
+ "s plit",
+ "eg o",
+ "e go",
+ "git hub",
+ "g ithub",
+ "▁) ,",
+ "▁ ),",
+ "RO M",
+ "R OM",
+ "▁ch air",
+ "▁cha ir",
+ "▁ chair",
+ "▁t rade",
+ "▁tr ade",
+ "▁trad e",
+ "▁tra de",
+ "▁n icht",
+ "▁ni cht",
+ "▁nic ht",
+ "To p",
+ "T op",
+ "St ore",
+ "▁p arte",
+ "▁part e",
+ "▁par te",
+ "pro ject",
+ "ni a",
+ "n ia",
+ "▁в ід",
+ "▁ві д",
+ "wa r",
+ "w ar",
+ "▁Pro f",
+ "▁Pr of",
+ "▁c aught",
+ "Th read",
+ "ст ва",
+ "ств а",
+ "с тва",
+ "aut hor",
+ "auth or",
+ "▁d oll",
+ "▁do ll",
+ "▁dol l",
+ "▁h arm",
+ "▁ha rm",
+ "▁har m",
+ "▁ harm",
+ "▁G en",
+ "▁Ge n",
+ "▁ Gen",
+ "tr ee",
+ "tre e",
+ "t ree",
+ "et ime",
+ "eti me",
+ "e time",
+ "cf g",
+ "c fg",
+ "▁gu ys",
+ "▁guy s",
+ "▁Cal ifornia",
+ "▁G reen",
+ "▁Gr een",
+ "▁Gre en",
+ "▁Gree n",
+ "▁ Green",
+ "▁mov ement",
+ "▁move ment",
+ "▁mo vement",
+ "ie j",
+ "i ej",
+ "▁stat ement",
+ "▁state ment",
+ "▁ statement",
+ "▁se eing",
+ "▁see ing",
+ "▁h aven",
+ "▁have n",
+ "▁ha ven",
+ "▁hav en",
+ "vent ion",
+ "v ention",
+ "S L",
+ "ched ul",
+ "ie rt",
+ "ier t",
+ "i ert",
+ "▁pr imary",
+ "▁prim ary",
+ "▁pri mary",
+ "▁prima ry",
+ "▁ primary",
+ "▁c ivil",
+ "▁ci vil",
+ "▁civ il",
+ "ri an",
+ "ria n",
+ "r ian",
+ "▁b utton",
+ "▁but ton",
+ "▁butt on",
+ "▁ button",
+ "▁l ived",
+ "▁li ved",
+ "▁live d",
+ "▁liv ed",
+ "P ass",
+ "so r",
+ "s or",
+ "▁watch ing",
+ "▁wat ching",
+ "▁sk ills",
+ "▁skill s",
+ "te e",
+ "t ee",
+ "Le vel",
+ "L evel",
+ "▁sc ient",
+ "h s",
+ "▁a gre",
+ "▁ag re",
+ "ca t",
+ "c at",
+ "▁t end",
+ "▁te nd",
+ "▁ten d",
+ "▁M ill",
+ "▁Mil l",
+ "▁Mi ll",
+ "▁ Mill",
+ "▁C ap",
+ "▁Ca p",
+ "▁ Cap",
+ "OR D",
+ "O RD",
+ "gl e",
+ "g le",
+ "▁с во",
+ "» ,",
+ "▁a head",
+ "▁ah ead",
+ "ve st",
+ "ves t",
+ "v est",
+ "▁J ose",
+ "▁Jo se",
+ "▁Jos e",
+ "is cher",
+ "isch er",
+ "ische r",
+ "isc her",
+ "ș i",
+ "▁le aving",
+ "▁д ля",
+ "▁s outh",
+ "▁so uth",
+ "▁sou th",
+ "▁sout h",
+ "▁con sum",
+ "▁cons um",
+ "▁ consum",
+ "R ange",
+ "▁activ ities",
+ "Se c",
+ "S ec",
+ "▁s ales",
+ "▁sa les",
+ "▁sal es",
+ "▁sale s",
+ "▁f ix",
+ "▁fi x",
+ "▁ fix",
+ "▁j ed",
+ "▁je d",
+ "▁ jed",
+ "ru m",
+ "r um",
+ "ve ctor",
+ "vec tor",
+ "v ector",
+ "▁s pot",
+ "▁sp ot",
+ "▁spo t",
+ "▁ spot",
+ "▁man ufact",
+ "к т",
+ "or row",
+ "orr ow",
+ "si gn",
+ "sig n",
+ "s ign",
+ "▁col lege",
+ "▁colle ge",
+ "▁colleg e",
+ "▁d river",
+ "▁dr iver",
+ "▁dri ver",
+ "▁driv er",
+ "▁drive r",
+ "▁ driver",
+ "▁def initely",
+ "▁definit ely",
+ "▁s pend",
+ "▁sp end",
+ "▁spe nd",
+ "miss ion",
+ "m ission",
+ "з у",
+ "at ively",
+ "ative ly",
+ "ativ ely",
+ "b i",
+ "Call back",
+ "▁particular ly",
+ "▁particul arly",
+ "▁h ell",
+ "▁he ll",
+ "▁hel l",
+ "▁ hell",
+ "▁p ool",
+ "▁po ol",
+ "▁ pool",
+ "PR E",
+ "P RE",
+ "▁cle arly",
+ "▁clear ly",
+ "P T",
+ "ot hes",
+ "oth es",
+ "othe s",
+ "▁I d",
+ "▁ Id",
+ "Loc ation",
+ "L ocation",
+ "▁R un",
+ "▁Ru n",
+ "▁ Run",
+ "▁f ixed",
+ "▁fix ed",
+ "▁ fixed",
+ "▁H and",
+ "▁Ha nd",
+ "▁Han d",
+ "▁ Hand",
+ "ba l",
+ "b al",
+ "d ouble",
+ "C an",
+ "Om ega",
+ "▁chall eng",
+ "▁stand ing",
+ "▁stan ding",
+ "▁ standing",
+ "it en",
+ "ite n",
+ "i ten",
+ "▁me chan",
+ "▁d urch",
+ "▁dur ch",
+ "▁d ell",
+ "▁de ll",
+ "▁del l",
+ "▁rais ed",
+ "▁raise d",
+ "▁ra ised",
+ "▁we ak",
+ "▁ weak",
+ "▁D u",
+ "▁ Du",
+ "gr ad",
+ "gra d",
+ "g rad",
+ "▁sc ene",
+ "▁scen e",
+ "▁ scene",
+ "pos s",
+ "po ss",
+ "p oss",
+ "▁t on",
+ "▁to n",
+ "▁ ton",
+ "▁e arth",
+ "▁ear th",
+ "ul ations",
+ "ulation s",
+ "▁str ength",
+ "▁stre ngth",
+ "▁streng th",
+ "ak ed",
+ "ake d",
+ "a ked",
+ "▁re main",
+ "▁rem ain",
+ "▁B i",
+ "▁ Bi",
+ "▁custom er",
+ "▁cust omer",
+ "▁ customer",
+ "ran ge",
+ "r ange",
+ "▁inter ested",
+ "▁interest ed",
+ "ON E",
+ "O NE",
+ "▁c off",
+ "▁co ff",
+ "re quire",
+ "requ ire",
+ "▁On ly",
+ "▁ Only",
+ "▁W eb",
+ "▁We b",
+ "▁ Web",
+ "▁f arm",
+ "▁far m",
+ "▁fa rm",
+ "▁act ivity",
+ "▁activ ity",
+ "▁ activity",
+ "▁r out",
+ "▁ro ut",
+ "▁rou t",
+ "bl ing",
+ "b ling",
+ "S Y",
+ "▁Rich ard",
+ "▁Ric hard",
+ "▁R ef",
+ "▁Re f",
+ "▁ Ref",
+ "▁ко н",
+ "▁к он",
+ "▁ кон",
+ "▁j un",
+ "▁ju n",
+ "bo rn",
+ "bor n",
+ "b orn",
+ "ij n",
+ "Config uration",
+ "um an",
+ "uma n",
+ "u man",
+ "E E",
+ "▁mar ried",
+ "▁З а",
+ "▁ За",
+ "▁f at",
+ "▁fa t",
+ "▁k id",
+ "▁ki d",
+ "▁T ur",
+ "▁Tu r",
+ "▁ Tur",
+ "▁off ered",
+ "▁offer ed",
+ "ni c",
+ "n ic",
+ "▁B ig",
+ "▁Bi g",
+ "▁ Big",
+ "Ga mma",
+ "G amma",
+ "▁He alth",
+ "▁ Health",
+ "▁T R",
+ "▁ TR",
+ "▁s ię",
+ "▁si ę",
+ "▁const ruction",
+ "▁construct ion",
+ "▁constr uction",
+ "▁constru ction",
+ "▁ construction",
+ "▁Ch urch",
+ "▁B et",
+ "▁Be t",
+ "▁ Bet",
+ "bu s",
+ "b us",
+ "▁e arn",
+ "▁ear n",
+ "ri ct",
+ "ric t",
+ "r ict",
+ "▁п ра",
+ "▁пр а",
+ "▁ пра",
+ "▁br ain",
+ "▁bra in",
+ "▁f ra",
+ "▁fr a",
+ "▁O p",
+ "▁ Op",
+ "FI G",
+ "F IG",
+ "em a",
+ "e ma",
+ "▁Europe an",
+ "▁S aint",
+ "▁Sa int",
+ "▁ Saint",
+ "AR E",
+ "A RE",
+ "ur i",
+ "u ri",
+ "▁R iver",
+ "{ }",
+ "▁s itting",
+ "▁sit ting",
+ "▁under standing",
+ "▁understand ing",
+ "▁pl ans",
+ "▁plan s",
+ "rop ri",
+ "▁old er",
+ "▁ol der",
+ "▁ older",
+ "▁pres sure",
+ "▁press ure",
+ "Im pl",
+ "Imp l",
+ "▁pe ace",
+ "Conne ction",
+ "Conn ection",
+ "Connect ion",
+ "▁f i",
+ "▁ fi",
+ "ri ch",
+ "ric h",
+ "r ich",
+ "▁sh ut",
+ "ap ers",
+ "ape rs",
+ "aper s",
+ "a pers",
+ "Po rt",
+ "P ort",
+ "▁L ook",
+ "▁Lo ok",
+ "▁ Look",
+ "ri m",
+ "r im",
+ "au th",
+ "aut h",
+ "a uth",
+ "au to",
+ "aut o",
+ "a uto",
+ "▁high ly",
+ "▁un less",
+ "▁W al",
+ "▁Wa l",
+ "▁re n",
+ "▁r en",
+ "▁ ren",
+ "w s",
+ "▁c ore",
+ "▁co re",
+ "▁cor e",
+ "▁ core",
+ "( -",
+ "▁c lim",
+ "▁cl im",
+ "ru it",
+ "r uit",
+ "▁call back",
+ "▁ callback",
+ "he st",
+ "hes t",
+ "h est",
+ "▁Char les",
+ "▁Charl es",
+ "▁L ong",
+ "▁Lo ng",
+ "▁ Long",
+ "} =",
+ "ъ р",
+ "▁sh ared",
+ "▁share d",
+ "▁shar ed",
+ "▁sha red",
+ "▁ shared",
+ "ul ated",
+ "ula ted",
+ "ulate d",
+ "gorith m",
+ "▁H ome",
+ "▁Ho me",
+ "▁Hom e",
+ "▁ Home",
+ "▁vill age",
+ "▁vil lage",
+ "ee s",
+ "e es",
+ "s v",
+ "▁rest aur",
+ "re y",
+ "r ey",
+ "▁C ast",
+ "▁Cas t",
+ "▁Ca st",
+ "▁ Cast",
+ "▁P erson",
+ "▁Per son",
+ "▁Pers on",
+ "▁ Person",
+ "ки й",
+ "▁organ iz",
+ "▁R ad",
+ "▁Ra d",
+ "▁ Rad",
+ "pon ents",
+ "ponent s",
+ "▁wer den",
+ "▁werd en",
+ "▁b ow",
+ "▁bo w",
+ "▁ bow",
+ "se n",
+ "s en",
+ "am i",
+ "a mi",
+ "Inter face",
+ "▁b asis",
+ "▁bas is",
+ "▁ba sis",
+ "▁Comp any",
+ "▁Compan y",
+ "▁ Company",
+ "er nel",
+ "ern el",
+ "erne l",
+ "it u",
+ "i tu",
+ "Has h",
+ "Ha sh",
+ "H ash",
+ "▁a an",
+ "▁ х",
+ "▁s mile",
+ "▁sm ile",
+ "x ml",
+ "▁s cen",
+ "▁sc en",
+ "am m",
+ "a mm",
+ "to ol",
+ "too l",
+ "t ool",
+ "ar ia",
+ "ari a",
+ "a ria",
+ "▁acc ur",
+ "▁ac cur",
+ "▁ accur",
+ "set tings",
+ "setting s",
+ "▁Jes us",
+ "ac ement",
+ "ace ment",
+ "po wer",
+ "pow er",
+ "p ower",
+ "( !",
+ "▁c alls",
+ "▁call s",
+ "▁cal ls",
+ "▁ calls",
+ "▁bas ic",
+ "▁ basic",
+ "▁set tings",
+ "▁sett ings",
+ "▁setting s",
+ "▁ settings",
+ "ri pt",
+ "rip t",
+ "r ipt",
+ "po ol",
+ "p ool",
+ "ct ors",
+ "ctor s",
+ "▁Found ation",
+ "▁ Foundation",
+ "▁we ap",
+ "KE Y",
+ "K EY",
+ "fo ot",
+ "foo t",
+ "f oot",
+ "▁r adio",
+ "▁rad io",
+ "▁radi o",
+ "▁ radio",
+ "▁hel ped",
+ "▁help ed",
+ "ma nn",
+ "man n",
+ "m ann",
+ "▁j ump",
+ "▁ju mp",
+ "▁t ick",
+ "▁ti ck",
+ "▁ tick",
+ "▁gr owing",
+ "▁grow ing",
+ "▁gro wing",
+ "at en",
+ "ate n",
+ "a ten",
+ "re al",
+ "rea l",
+ "▁incre asing",
+ "Dev ice",
+ "var epsilon",
+ "vare psilon",
+ "▁s ets",
+ "▁se ts",
+ "▁set s",
+ "▁ sets",
+ "▁adv ant",
+ "Op en",
+ "O pen",
+ "▁re asons",
+ "▁reason s",
+ "▁sup posed",
+ "▁supp osed",
+ "▁suppose d",
+ "oe s",
+ "o es",
+ "ed e",
+ "e de",
+ "te en",
+ "tee n",
+ "t een",
+ "if def",
+ "▁de lete",
+ "▁del ete",
+ "▁delet e",
+ "▁ delete",
+ "▁& =",
+ "▁ &=",
+ "▁B ill",
+ "▁Bi ll",
+ "▁Bil l",
+ "▁ Bill",
+ "▁a im",
+ "▁ai m",
+ "▁ aim",
+ "▁O k",
+ "▁ Ok",
+ "▁A v",
+ "▁ Av",
+ "re ci",
+ "rec i",
+ "ac ks",
+ "ack s",
+ "a cks",
+ "is te",
+ "ist e",
+ "i ste",
+ "Pro perties",
+ "▁t mp",
+ "▁tm p",
+ "▁ tmp",
+ "▁d ei",
+ "▁de i",
+ "PE R",
+ "P ER",
+ "D C",
+ "st a",
+ "s ta",
+ "ни и",
+ "▁lim ited",
+ "▁limit ed",
+ "▁ limited",
+ "▁great er",
+ "▁gre ater",
+ "de scription",
+ "des cription",
+ "or i",
+ "o ri",
+ "ain ts",
+ "aint s",
+ "▁h y",
+ "▁ hy",
+ "▁M el",
+ "▁Me l",
+ "▁C H",
+ "▁ CH",
+ "con s",
+ "co ns",
+ "c ons",
+ "▁sur round",
+ "▁W ho",
+ "▁Wh o",
+ "▁ Who",
+ "ar c",
+ "a rc",
+ "▁te lev",
+ "▁tele v",
+ "▁tel ev",
+ "it ution",
+ "itut ion",
+ "▁e qual",
+ "▁equ al",
+ "▁eq ual",
+ "▁ equal",
+ "к і",
+ "▁Is rael",
+ "ä h",
+ "▁C aption",
+ "▁Capt ion",
+ "▁Ca ption",
+ "▁ex erc",
+ "em por",
+ "emp or",
+ "▁+ +",
+ "▁ ++",
+ "▁l ib",
+ "▁li b",
+ "▁ lib",
+ "ma ke",
+ "m ake",
+ "▁M A",
+ "▁ MA",
+ "co py",
+ "cop y",
+ "c opy",
+ "f riend",
+ "▁ко то",
+ "▁ кото",
+ "▁dam age",
+ "▁\\ ,",
+ "▁ \\,",
+ "od ed",
+ "ode d",
+ "o ded",
+ "▁n one",
+ "▁no ne",
+ "▁non e",
+ "▁ none",
+ "▁ev alu",
+ "▁eval u",
+ "▁ evalu",
+ "st on",
+ "sto n",
+ "s ton",
+ "> ,",
+ "FO R",
+ "F OR",
+ "▁n orm",
+ "▁no rm",
+ "▁nor m",
+ "▁ norm",
+ "ap pe",
+ "app e",
+ "a ppe",
+ "S ession",
+ "▁ad ult",
+ "▁h ospital",
+ "▁hosp ital",
+ "▁recomm end",
+ "pro perty",
+ "ste in",
+ "fin al",
+ "fi nal",
+ "f inal",
+ "▁n u",
+ "▁ nu",
+ "se cond",
+ "sec ond",
+ "▁a spect",
+ "▁as pect",
+ "▁asp ect",
+ "\") ]",
+ "\" )]",
+ "же н",
+ "ж ен",
+ "am ento",
+ "ament o",
+ "amen to",
+ "▁r ac",
+ "▁ra c",
+ "▁ rac",
+ "sa ve",
+ "s ave",
+ "▁foot ball",
+ "A b",
+ "un gs",
+ "ung s",
+ "ab il",
+ "abi l",
+ "a bil",
+ "▁Ar ch",
+ "▁Arc h",
+ "▁ Arch",
+ "sys tem",
+ "s ystem",
+ "hi st",
+ "his t",
+ "h ist",
+ "▁l uck",
+ "▁lu ck",
+ "▁luc k",
+ "re nder",
+ "ren der",
+ "rend er",
+ "r ender",
+ "▁se in",
+ "▁sei n",
+ "ion i",
+ "io ni",
+ "i oni",
+ "▁r ot",
+ "▁ro t",
+ "▁ rot",
+ "▁cor ner",
+ "▁corn er",
+ "▁app ropri",
+ "▁ap propri",
+ "▁ appropri",
+ "▁Soft ware",
+ "▁t ele",
+ "▁te le",
+ "▁tel e",
+ "▁ tele",
+ "De lete",
+ "Dele te",
+ "Del ete",
+ "▁Acc ording",
+ "▁pr ison",
+ "▁pri son",
+ "▁ prison",
+ "▁l ic",
+ "▁li c",
+ "▁ lic",
+ "▁м и",
+ "▁ ми",
+ "ter m",
+ "te rm",
+ "t erm",
+ "se ts",
+ "set s",
+ "s ets",
+ "▁v el",
+ "▁ve l",
+ "▁ vel",
+ "▁r ank",
+ "▁ran k",
+ "▁ rank",
+ "▁ex isting",
+ "▁exist ing",
+ "▁ existing",
+ "▁V ir",
+ "▁Vi r",
+ "▁t rip",
+ "▁tr ip",
+ "▁tri p",
+ "▁м у",
+ "▁ му",
+ "av ax",
+ "ava x",
+ "▁r is",
+ "▁ri s",
+ "▁ ris",
+ "▁def ine",
+ "▁defin e",
+ "▁ define",
+ "▁he at",
+ "ca r",
+ "c ar",
+ "▁con vert",
+ "▁conv ert",
+ "▁conver t",
+ "▁conve rt",
+ "▁ convert",
+ "em ail",
+ "ema il",
+ "e mail",
+ "▁U nder",
+ "▁Un der",
+ "▁Und er",
+ "▁ Under",
+ "▁ Ш",
+ "▁G rand",
+ "▁Gr and",
+ "▁Gran d",
+ "▁Gra nd",
+ "▁ex ists",
+ "▁exist s",
+ "▁ exists",
+ "sy s",
+ "s ys",
+ "ef f",
+ "e ff",
+ "▁T op",
+ "▁To p",
+ "▁ Top",
+ "▁ č",
+ "▁t empor",
+ "▁tem por",
+ "▁temp or",
+ "▁tempo r",
+ "▁arg uments",
+ "▁argument s",
+ "▁ arguments",
+ "▁support ed",
+ "▁supp orted",
+ "▁ supported",
+ "en sed",
+ "ens ed",
+ "ense d",
+ "▁Franc is",
+ "▁co ord",
+ "▁ coord",
+ "▁achie ve",
+ "▁N ame",
+ "▁Na me",
+ "▁Nam e",
+ "▁ Name",
+ "▁J ahr",
+ "▁Jah r",
+ "▁Ja hr",
+ "▁G i",
+ "sh e",
+ "s he",
+ "▁D ev",
+ "▁De v",
+ "▁ Dev",
+ "▁a lla",
+ "▁al la",
+ "▁all a",
+ "▁ alla",
+ "▁W IT",
+ "ag ment",
+ "c ustom",
+ "al ls",
+ "all s",
+ "& &",
+ "W E",
+ "▁h olding",
+ "▁hold ing",
+ "▁hol ding",
+ "pro totype",
+ "proto type",
+ "prot otype",
+ "▁f ing",
+ "▁fin g",
+ "▁fi ng",
+ "▁b ag",
+ "▁ba g",
+ "▁ bag",
+ "▁Par ty",
+ "▁Part y",
+ "st ack",
+ "sta ck",
+ "▁econom ic",
+ "▁G al",
+ "▁Ga l",
+ "id ents",
+ "ident s",
+ "iden ts",
+ "▁J un",
+ "▁Ju n",
+ "▁sh owed",
+ "▁show ed",
+ "os h",
+ "o sh",
+ "▁B ay",
+ "▁Ba y",
+ "▁ Bay",
+ "ma il",
+ "m ail",
+ "▁S O",
+ "▁ SO",
+ "▁\" <",
+ "graph ics",
+ "▁f u",
+ "▁ fu",
+ "cl ick",
+ "cli ck",
+ "c lick",
+ "▁b attle",
+ "▁batt le",
+ "▁bat tle",
+ "{ {",
+ "▁E vent",
+ "▁Even t",
+ "▁Ev ent",
+ "▁Eve nt",
+ "▁ Event",
+ "ri or",
+ "rio r",
+ "r ior",
+ "ch aft",
+ "cha ft",
+ "▁f avorite",
+ "▁favor ite",
+ "us ive",
+ "sup port",
+ "supp ort",
+ "s upport",
+ "b m",
+ "K ind",
+ "▁saf ety",
+ "▁safe ty",
+ "▁E nt",
+ "▁En t",
+ "▁ Ent",
+ "cu p",
+ "c up",
+ "▁Austral ia",
+ "▁dest roy",
+ "▁destro y",
+ "▁ destroy",
+ "▁organ ization",
+ "▁organiz ation",
+ "id en",
+ "ide n",
+ "i den",
+ "######## ########",
+ "de c",
+ "d ec",
+ "▁z a",
+ "▁ za",
+ "▁s even",
+ "▁se ven",
+ "▁ seven",
+ "ar ely",
+ "are ly",
+ "arel y",
+ "▁f lag",
+ "▁fl ag",
+ "▁ flag",
+ "Di r",
+ "D ir",
+ "▁C arl",
+ "▁Car l",
+ "▁Ca rl",
+ "▁do ctor",
+ "▁doc tor",
+ "▁var iety",
+ "▁vari ety",
+ "▁L in",
+ "▁Li n",
+ "▁ Lin",
+ "▁t om",
+ "▁to m",
+ "▁ tom",
+ "^{ (",
+ "^ {(",
+ "B o",
+ "an tes",
+ "ant es",
+ "ante s",
+ "▁m ine",
+ "▁min e",
+ "▁mi ne",
+ "▁ mine",
+ "▁M it",
+ "▁Mi t",
+ "▁de scribe",
+ "▁desc ribe",
+ "▁describ e",
+ "Ar gs",
+ "Arg s",
+ "L S",
+ "AP I",
+ "A PI",
+ "▁L uc",
+ "▁Lu c",
+ "▁ Luc",
+ "ph one",
+ "▁sc ience",
+ "▁ science",
+ "▁O per",
+ "▁Op er",
+ "▁ Oper",
+ "Ne xt",
+ "N ext",
+ "▁invest ig",
+ "▁demon str",
+ "▁G overn",
+ "▁Go vern",
+ "▁object s",
+ "▁ objects",
+ "▁Lou is",
+ "▁Lo uis",
+ "▁Return s",
+ "▁ Returns",
+ "▁h an",
+ "▁ha n",
+ "▁ han",
+ "na m",
+ "n am",
+ "▁com me",
+ "▁comm e",
+ "▁pres ence",
+ "▁p el",
+ "▁pe l",
+ "▁ pel",
+ "▁det ect",
+ "▁ detect",
+ ") =",
+ "▁Ch inese",
+ "▁r ich",
+ "▁ri ch",
+ "▁ric h",
+ "▁ rich",
+ "▁class es",
+ "▁classe s",
+ "▁clas ses",
+ "▁ classes",
+ "▁exp and",
+ "▁ expand",
+ "▁D om",
+ "▁Do m",
+ "▁ Dom",
+ "▁D ec",
+ "▁De c",
+ "▁ Dec",
+ "s n",
+ "pe ed",
+ "p eed",
+ "▁J im",
+ "▁Ji m",
+ "sh ould",
+ "▁Sm ith",
+ "▁p ages",
+ "▁page s",
+ "▁pa ges",
+ "▁pag es",
+ "▁ pages",
+ "▁Je an",
+ "ri cs",
+ "ric s",
+ "r ics",
+ "▁S und",
+ "▁Su nd",
+ "▁Sun d",
+ "ad s",
+ "a ds",
+ "▁The ir",
+ "un icip",
+ "uni cip",
+ "unic ip",
+ "в у",
+ "▁down load",
+ "▁ download",
+ "▁st ress",
+ "▁str ess",
+ "▁stre ss",
+ "▁P et",
+ "▁Pe t",
+ "▁ Pet",
+ "me nu",
+ "men u",
+ "m enu",
+ "re me",
+ "rem e",
+ "r eme",
+ "▁com pared",
+ "▁comp ared",
+ "▁compar ed",
+ "▁compare d",
+ "St e",
+ "S te",
+ "IN D",
+ "I ND",
+ "cont ainer",
+ "▁Ind ian",
+ "▁India n",
+ "or en",
+ "ore n",
+ "o ren",
+ "▁s es",
+ "▁se s",
+ "▁ ses",
+ "▁W he",
+ "▁Wh e",
+ "▁ Whe",
+ "▁r oku",
+ "▁ro ku",
+ "▁estab lished",
+ "▁establish ed",
+ "▁gener ally",
+ "▁general ly",
+ "▁f le",
+ "▁fl e",
+ "__ (",
+ "_ _(",
+ "=\" +",
+ "= \"+",
+ "V ar",
+ "▁M ake",
+ "▁Ma ke",
+ "▁Mak e",
+ "▁ Make",
+ "▁rem oved",
+ "▁remove d",
+ "▁ removed",
+ "z z",
+ "ü n",
+ "▁m ix",
+ "▁mi x",
+ "▁ mix",
+ "er k",
+ "iat ion",
+ "i ation",
+ "ou ter",
+ "out er",
+ "oute r",
+ "o uter",
+ "S K",
+ "▁be comes",
+ "▁bec omes",
+ "▁become s",
+ "▁H all",
+ "▁Ha ll",
+ "▁Hal l",
+ "sc ious",
+ "▁w atched",
+ "▁watch ed",
+ "▁wat ched",
+ "▁g ather",
+ "▁ga ther",
+ "▁ gather",
+ "▁Res ult",
+ "▁ Result",
+ "pro of",
+ "pa y",
+ "p ay",
+ "▁produ ced",
+ "▁produce d",
+ "▁prod uced",
+ "▁| =",
+ "▁b order",
+ "▁bord er",
+ "▁bor der",
+ "▁ border",
+ "▁d in",
+ "▁di n",
+ "▁s cript",
+ "▁sc ript",
+ "▁scr ipt",
+ "▁ script",
+ "▁a ctions",
+ "▁act ions",
+ "▁action s",
+ "▁ actions",
+ "▁m as",
+ "▁ma s",
+ "▁ mas",
+ "щ а",
+ "oot h",
+ "oo th",
+ "o oth",
+ "▁Te chn",
+ "▁Tech n",
+ "Js on",
+ "J son",
+ "▁f illed",
+ "▁fil led",
+ "▁fill ed",
+ "▁ filled",
+ "де н",
+ "д ен",
+ "und le",
+ "ст у",
+ "с ту",
+ "To ol",
+ "Too l",
+ "T ool",
+ "▁k ing",
+ "▁ki ng",
+ "▁kin g",
+ "▁ king",
+ "▁v en",
+ "▁ve n",
+ "▁ ven",
+ "st ra",
+ "str a",
+ "s tra",
+ "▁pre dict",
+ "▁pred ict",
+ "▁ predict",
+ "▁l ui",
+ "▁lu i",
+ "▁WAR RAN",
+ "▁F un",
+ "▁Fu n",
+ "▁ Fun",
+ "Sc ript",
+ "S cript",
+ "▁power ful",
+ "▁l ose",
+ "▁lo se",
+ "▁los e",
+ "at ically",
+ "atic ally",
+ "▁d aily",
+ "▁da ily",
+ "▁dai ly",
+ "▁r ing",
+ "▁ri ng",
+ "▁ ring",
+ "▁ar rived",
+ "▁arriv ed",
+ "▁arr ived",
+ "▁arrive d",
+ "St ack",
+ "sc ope",
+ "s cope",
+ "▁B ack",
+ "▁Ba ck",
+ "▁ Back",
+ "el ij",
+ "eli j",
+ "e lij",
+ "▁z e",
+ "▁ ze",
+ "ke ys",
+ "key s",
+ "{ \"",
+ "VI D",
+ "V ID",
+ "▁l icense",
+ "▁lic ense",
+ "▁ license",
+ "wh at",
+ "w hat",
+ "▁pro ced",
+ "▁proc ed",
+ "ra nt",
+ "ran t",
+ "r ant",
+ "est ival",
+ "ag ram",
+ "agr am",
+ "agra m",
+ "a gram",
+ "▁L O",
+ "▁ LO",
+ "▁Hen ry",
+ "▁fl ags",
+ "▁flag s",
+ "▁ flags",
+ "Do wn",
+ "D own",
+ "scri ption",
+ "script ion",
+ "s cription",
+ "▁famil ies",
+ "▁familie s",
+ "is se",
+ "iss e",
+ "bo ur",
+ "b our",
+ "▁B ur",
+ "▁Bu r",
+ "— \"",
+ "▁b rief",
+ "▁br ief",
+ "▁ brief",
+ "▁cre ating",
+ "▁creat ing",
+ "▁cl ients",
+ "▁client s",
+ "ran gle",
+ "r angle",
+ "▁amaz ing",
+ "▁s ind",
+ "▁si nd",
+ "▁sin d",
+ "▁cover ed",
+ "▁cov ered",
+ "▁ covered",
+ "We ll",
+ "W ell",
+ "ст е",
+ "с те",
+ "то р",
+ "т ор",
+ "▁B as",
+ "▁Ba s",
+ "▁ Bas",
+ "to tal",
+ "tot al",
+ "t otal",
+ "▁I nit",
+ "▁In it",
+ "▁ Init",
+ "▁s and",
+ "▁sa nd",
+ "▁san d",
+ "Un it",
+ "U nit",
+ "▁mur der",
+ "▁b right",
+ "▁br ight",
+ "▁brig ht",
+ "▁t rav",
+ "▁tr av",
+ "▁tra v",
+ "ic ans",
+ "ica ns",
+ "ican s",
+ "▁att ribute",
+ "▁attribut e",
+ "▁ attribute",
+ "f c",
+ "▁pl aced",
+ "▁place d",
+ "▁plac ed",
+ "ES T",
+ "E ST",
+ "Var i",
+ "V ari",
+ "▁c os",
+ "▁co s",
+ "▁ cos",
+ "▁at tract",
+ "▁att ract",
+ "▁attr act",
+ "▁attra ct",
+ "an el",
+ "ane l",
+ "a nel",
+ "}) .",
+ "} ).",
+ "by tes",
+ "byte s",
+ "▁p arse",
+ "▁par se",
+ "▁ parse",
+ "▁be long",
+ "▁bel ong",
+ "B N",
+ "▁S ol",
+ "▁So l",
+ "P o",
+ "` ,",
+ "▁c alling",
+ "▁call ing",
+ "▁cal ling",
+ "▁? >",
+ "▁ ?>",
+ "▁it er",
+ "▁i ter",
+ "▁ iter",
+ "▁u rl",
+ "▁ur l",
+ "▁ url",
+ "▁ev ening",
+ "▁even ing",
+ "re ek",
+ "ree k",
+ "▁hon est",
+ "▁direct or",
+ "▁dire ctor",
+ "▁dir ector",
+ "R C",
+ "▁s olid",
+ "▁sol id",
+ "▁ solid",
+ "▁ph il",
+ "ie ne",
+ "ien e",
+ "i ene",
+ "FA ULT",
+ "co pe",
+ "cop e",
+ "c ope",
+ "▁Hist ory",
+ "▁Histor y",
+ "▁Hi story",
+ "▁ History",
+ "▁Te am",
+ "▁ Team",
+ "ree dom",
+ "reed om",
+ "▁r u",
+ "▁ ru",
+ "U B",
+ "▁w orse",
+ "▁wor se",
+ "im o",
+ "i mo",
+ "Ma t",
+ "M at",
+ "▁M ex",
+ "▁Me x",
+ "ac tor",
+ "act or",
+ "a ctor",
+ "▁v or",
+ "▁vo r",
+ "▁ vor",
+ "ть ся",
+ "▁exper iment",
+ "▁experi ment",
+ "▁P lay",
+ "▁Pl ay",
+ "▁ Play",
+ "▁An other",
+ "▁happ ens",
+ "▁happen s",
+ "ua n",
+ "u an",
+ "▁pat ients",
+ "▁patient s",
+ "▁re nd",
+ "▁r end",
+ "▁ren d",
+ "▁ rend",
+ "▁M o",
+ "▁ Mo",
+ "▁T ex",
+ "▁Te x",
+ "▁ Tex",
+ "▁w ed",
+ "▁we d",
+ "▁ wed",
+ "t n",
+ "in sert",
+ "ins ert",
+ "▁п а",
+ "▁ па",
+ "▁an ti",
+ "▁ant i",
+ "▁ anti",
+ "Mat ch",
+ "M atch",
+ "ampions hip",
+ "ampion ship",
+ "▁for ces",
+ "▁force s",
+ "▁H ot",
+ "▁Ho t",
+ "▁ Hot",
+ "▁ph ase",
+ "▁ phase",
+ "▁t emplate",
+ "▁templ ate",
+ "▁temp late",
+ "▁ template",
+ "st op",
+ "sto p",
+ "s top",
+ "ic ated",
+ "ica ted",
+ "icate d",
+ "▁man aged",
+ "▁manage d",
+ "▁ managed",
+ "wa it",
+ "w ait",
+ "▁* (",
+ "▁ *(",
+ "G B",
+ "▁app oint",
+ "▁ap point",
+ "▁ appoint",
+ "ł a",
+ "▁s tick",
+ "▁st ick",
+ "▁ stick",
+ "▁F OR",
+ "▁FO R",
+ "▁ FOR",
+ "▁V is",
+ "▁Vi s",
+ "▁ Vis",
+ "to r",
+ "t or",
+ "▁p ř",
+ "qu est",
+ "que st",
+ "ques t",
+ "q uest",
+ "us es",
+ "use s",
+ "u ses",
+ "\"); \r",
+ "\") ;\r",
+ "\" );\r",
+ "▁sudden ly",
+ "▁sud denly",
+ "é c",
+ "N D",
+ "ur op",
+ "uro p",
+ "u rop",
+ "ре д",
+ "▁ins urance",
+ "ac cess",
+ "acc ess",
+ "a ccess",
+ "un finished",
+ "▁t amb",
+ "▁ta mb",
+ "▁tam b",
+ "▁s ac",
+ "▁sa c",
+ "▁C ourt",
+ "▁Co urt",
+ "▁Cour t",
+ "▁Cou rt",
+ "▁miss ing",
+ "▁mis sing",
+ "▁ missing",
+ "▁W here",
+ "▁Wh ere",
+ "▁Whe re",
+ "▁ Where",
+ "▁S um",
+ "▁Su m",
+ "▁ Sum",
+ "}^ {\\",
+ "}^{ \\",
+ "} ^{\\",
+ "▁s ua",
+ "▁su a",
+ "_ ,",
+ "▁th ick",
+ "▁Tr ump",
+ "▁Tru mp",
+ "▁oper ations",
+ "▁operation s",
+ "▁ operations",
+ "F S",
+ "▁de ux",
+ "d z",
+ "Temp late",
+ "T emplate",
+ "▁\" /",
+ "▁o dd",
+ "▁od d",
+ "▁ odd",
+ "▁re ality",
+ "▁real ity",
+ "▁te ams",
+ "▁team s",
+ "▁tea ms",
+ "▁c er",
+ "▁ce r",
+ "▁ cer",
+ "om a",
+ "o ma",
+ "▁ și",
+ "▁cl oud",
+ "▁clo ud",
+ "▁ cloud",
+ "▁Dep artment",
+ "N e",
+ "▁requ ires",
+ "▁require s",
+ "it ems",
+ "ite ms",
+ "item s",
+ "▁I II",
+ "▁II I",
+ "▁ III",
+ "right arrow",
+ ")- >",
+ ") ->",
+ "▁w riter",
+ "▁wr iter",
+ "▁writ er",
+ "▁write r",
+ "▁ writer",
+ "re place",
+ "rep lace",
+ "▁t hr",
+ "▁th r",
+ "je n",
+ "j en",
+ "▁o t",
+ "▁ ot",
+ "▁occ up",
+ "▁oc cup",
+ "▁ occup",
+ "▁event ually",
+ "▁M ath",
+ "▁Mat h",
+ "▁Ma th",
+ "▁ Math",
+ "▁con serv",
+ "▁cons erv",
+ "▁conse rv",
+ "am er",
+ "ame r",
+ "a mer",
+ "▁F ort",
+ "▁For t",
+ "▁Fo rt",
+ "▁d ry",
+ "▁dr y",
+ "▁sex ual",
+ "▁co sts",
+ "▁cost s",
+ "▁cos ts",
+ "▁for ms",
+ "▁form s",
+ "▁ forms",
+ "▁V ict",
+ "▁Vi ct",
+ "▁Vic t",
+ "PA R",
+ "P AR",
+ "frame work",
+ "▁д и",
+ "▁ ди",
+ "Oper ation",
+ "з на",
+ "wh ich",
+ "▁t ight",
+ "▁ti ght",
+ "In valid",
+ "▁part ner",
+ "▁п ред",
+ "▁пре д",
+ "▁th ank",
+ "▁than k",
+ "▁gu ard",
+ "▁ guard",
+ "he m",
+ "h em",
+ "Bo dy",
+ "B ody",
+ "▁e mot",
+ "▁em ot",
+ "I X",
+ "fa st",
+ "fas t",
+ "f ast",
+ "щ о",
+ "ñ o",
+ "ni ght",
+ "n ight",
+ "▁S ci",
+ "▁Sc i",
+ "ни ка",
+ "ник а",
+ "▁T O",
+ "▁ TO",
+ "▁individ uals",
+ "▁individual s",
+ "сс и",
+ "с си",
+ "}) ,",
+ "} ),",
+ "F alse",
+ "(\" %",
+ "( \"%",
+ "▁op tim",
+ "▁opt im",
+ "▁ optim",
+ "▁- ->",
+ "▁-- >",
+ "▁ -->",
+ "▁f actor",
+ "▁fact or",
+ "▁fac tor",
+ "▁fa ctor",
+ "▁ factor",
+ "▁sm aller",
+ "▁small er",
+ "▁con tain",
+ "▁cont ain",
+ "sp ect",
+ "spec t",
+ "spe ct",
+ "s pect",
+ "Eng ine",
+ "▁ann ounced",
+ "▁announ ced",
+ "▁announce d",
+ "▁Dem ocr",
+ "▁r ob",
+ "▁ro b",
+ "▁ rob",
+ "▁f lat",
+ "▁fl at",
+ "▁ flat",
+ "os oph",
+ "oso ph",
+ "Se arch",
+ "S earch",
+ "ah l",
+ "a hl",
+ "▁Ex ception",
+ "▁Except ion",
+ "▁ Exception",
+ "▁O l",
+ "equ als",
+ "eq uals",
+ "equal s",
+ "▁un ter",
+ "▁unt er",
+ "▁ unter",
+ "sh ape",
+ "sha pe",
+ "N S",
+ "Ob j",
+ "▁spec ies",
+ "▁spe cies",
+ "we ight",
+ "wei ght",
+ "w eight",
+ "yo u",
+ "y ou",
+ "▁e ste",
+ "▁est e",
+ "▁es te",
+ "▁ este",
+ "▁V iew",
+ "▁Vi ew",
+ "▁ View",
+ "▁m ission",
+ "▁miss ion",
+ "▁ mission",
+ "▁j ournal",
+ "▁jour nal",
+ "▁ journal",
+ "Value s",
+ "Val ues",
+ "▁ein em",
+ "▁eine m",
+ "is mo",
+ "ism o",
+ "▁project s",
+ "▁ projects",
+ "▁D as",
+ "▁Da s",
+ "ri ble",
+ "rib le",
+ "r ible",
+ "▁s erve",
+ "▁ser ve",
+ "▁serv e",
+ "▁ serve",
+ "▁op ening",
+ "▁open ing",
+ "▁h ur",
+ "▁program s",
+ "▁U SA",
+ "▁US A",
+ "▁ USA",
+ "il iar",
+ "ili ar",
+ "ilia r",
+ "id os",
+ "ido s",
+ "B r",
+ "est amp",
+ "esta mp",
+ "▁t ools",
+ "▁to ols",
+ "▁too ls",
+ "▁tool s",
+ "▁ tools",
+ "an ner",
+ "ann er",
+ "anne r",
+ "R T",
+ "▁St art",
+ "▁Star t",
+ "▁Sta rt",
+ "▁ Start",
+ "▁b ath",
+ "▁bat h",
+ "▁ba th",
+ "▁coff ee",
+ "or ter",
+ "ort er",
+ "orte r",
+ "in ternal",
+ "inter nal",
+ "intern al",
+ "file s",
+ "fil es",
+ "fi les",
+ "f iles",
+ "IN VAL",
+ "ak o",
+ "a ko",
+ "d t",
+ "▁Se cond",
+ "▁Sec ond",
+ "▁ Second",
+ "▁al loc",
+ "▁all oc",
+ "▁ alloc",
+ "▁en ded",
+ "▁end ed",
+ "▁ende d",
+ "▁ ended",
+ "ac ional",
+ "aci onal",
+ "acion al",
+ "acio nal",
+ "▁man ager",
+ "▁manage r",
+ "▁ manager",
+ "▁S un",
+ "▁Su n",
+ "▁ Sun",
+ "ag g",
+ "a gg",
+ "▁le ader",
+ "▁lead er",
+ "ol ved",
+ "olve d",
+ "olv ed",
+ "▁ч то",
+ "▁trad itional",
+ "▁tradition al",
+ "sh ot",
+ "s hot",
+ "ru p",
+ "r up",
+ "C F",
+ "▁E ach",
+ "▁ Each",
+ "w r",
+ "▁S om",
+ "▁So m",
+ "▁ Som",
+ "▁material s",
+ "▁mater ials",
+ "▁m sg",
+ "▁ms g",
+ "▁ msg",
+ "▁s yn",
+ "▁sy n",
+ "▁ syn",
+ "▁produ ce",
+ "▁prod uce",
+ "▁st orage",
+ "▁stor age",
+ "▁sto rage",
+ "▁ storage",
+ "sub section",
+ "▁S ie",
+ "▁Si e",
+ "▁I P",
+ "▁ IP",
+ "CE SS",
+ "▁w a",
+ "▁ wa",
+ "Re cord",
+ "Rec ord",
+ "▁mark eting",
+ "▁market ing",
+ "pl et",
+ "ple t",
+ "p let",
+ "D ialog",
+ "▁mention ed",
+ "▁ment ioned",
+ "▁N a",
+ "▁ Na",
+ "▁Un ion",
+ "▁ Union",
+ "▁A PI",
+ "▁AP I",
+ "▁ API",
+ "▁neg ative",
+ "▁ negative",
+ "tx t",
+ "t xt",
+ "▁eas ier",
+ "le gal",
+ "leg al",
+ "De p",
+ "D ep",
+ "▁no vel",
+ "▁nov el",
+ "▁nove l",
+ "eu r",
+ "e ur",
+ "ac ió",
+ "aci ó",
+ "a ció",
+ "▁B ud",
+ "▁Bu d",
+ "▁c arry",
+ "▁car ry",
+ "sch aft",
+ "s chaft",
+ "▁br oken",
+ "▁bro ken",
+ "▁broke n",
+ "▁t rees",
+ "▁tr ees",
+ "▁tre es",
+ "▁tree s",
+ ">( );",
+ ">() ;",
+ "> ();",
+ "▁e mb",
+ "▁em b",
+ "▁ emb",
+ "ie der",
+ "ied er",
+ "i eder",
+ "▁r oute",
+ "▁ro ute",
+ "▁rout e",
+ "▁rou te",
+ "▁ route",
+ "ik el",
+ "ike l",
+ "i kel",
+ "▁l isten",
+ "▁li sten",
+ "▁list en",
+ "▁ listen",
+ "ash ion",
+ "ashi on",
+ "▁M rs",
+ "▁Mr s",
+ "▁equip ment",
+ "ag ger",
+ "agg er",
+ "▁T hus",
+ "▁Th us",
+ "▁mat rix",
+ "▁ matrix",
+ "al la",
+ "all a",
+ "a lla",
+ "▁T our",
+ "▁To ur",
+ "▁con versation",
+ "▁convers ation",
+ "Mo n",
+ "M on",
+ "our nal",
+ "▁min ute",
+ "▁minut e",
+ "▁ minute",
+ "A m",
+ "Ap i",
+ "A pi",
+ "▁for get",
+ "▁forg et",
+ "M e",
+ "lev ant",
+ "te mp",
+ "tem p",
+ "t emp",
+ "▁t elling",
+ "▁tell ing",
+ "▁tel ling",
+ "mo ve",
+ "mov e",
+ "m ove",
+ "▁in dependent",
+ "▁independ ent",
+ "to String",
+ "ed it",
+ "edi t",
+ "e dit",
+ "▁J ac",
+ "▁Ja c",
+ "az z",
+ "a zz",
+ "re act",
+ "rea ct",
+ "▁c in",
+ "▁ci n",
+ "▁ cin",
+ "▁P rov",
+ "▁Pro v",
+ "▁Pr ov",
+ "▁ Prov",
+ "is ted",
+ "ist ed",
+ "iste d",
+ "i sted",
+ "▁h ash",
+ "▁has h",
+ "▁ha sh",
+ "▁ hash",
+ "on na",
+ "ik i",
+ "i ki",
+ "▁gener ated",
+ "▁generate d",
+ "▁gene rated",
+ "▁ generated",
+ "Re nder",
+ "Rend er",
+ "R ender",
+ "▁psy ch",
+ "▁ps ych",
+ "na v",
+ "n av",
+ "▁en tr",
+ "▁ent r",
+ "▁ entr",
+ "п ра",
+ "r x",
+ "AT H",
+ "A TH",
+ "▁ass ume",
+ "▁assum e",
+ "Tr ee",
+ "T ree",
+ "semb ly",
+ "sembl y",
+ "▁M att",
+ "▁Mat t",
+ "▁Ma tt",
+ "ca ption",
+ "c aption",
+ "▁s olutions",
+ "▁solution s",
+ "▁fa ith",
+ "▁fait h",
+ "▁dig ital",
+ "▁digit al",
+ "▁ex cell",
+ "▁exc ell",
+ "▁V ersion",
+ "▁Vers ion",
+ "▁ Version",
+ "De bug",
+ "D ebug",
+ "▁ж и",
+ "▁ жи",
+ "▁car ried",
+ "re set",
+ "res et",
+ "▁slow ly",
+ "an cing",
+ "anc ing",
+ "▁own er",
+ "▁ owner",
+ "▁T er",
+ "▁Te r",
+ "▁D id",
+ "▁Di d",
+ "▁ Did",
+ "▁g est",
+ "▁ge st",
+ "▁ges t",
+ "▁ gest",
+ "▁é té",
+ "▁ét é",
+ "▁ été",
+ "▁pro of",
+ "▁ proof",
+ "F ont",
+ "▁n ob",
+ "▁no b",
+ "▁ nob",
+ "C o",
+ "▁G NU",
+ "▁l iber",
+ "▁li ber",
+ "▁lib er",
+ "it ness",
+ "▁h ij",
+ "▁hi j",
+ "▁v ert",
+ "▁ver t",
+ "▁ve rt",
+ "▁ vert",
+ "ш а",
+ "FL AG",
+ "ME NT",
+ "M ENT",
+ "▁S on",
+ "▁So n",
+ "Mu lt",
+ "M ult",
+ "▁d istrict",
+ "▁di strict",
+ "▁dist rict",
+ "conne ct",
+ "conn ect",
+ "ject ion",
+ "je ction",
+ "j ection",
+ "ly mp",
+ "▁real ized",
+ "▁realize d",
+ "▁realiz ed",
+ "mo s",
+ "m os",
+ "y e",
+ "▁re nder",
+ "▁r ender",
+ "▁ren der",
+ "▁rend er",
+ "▁ render",
+ "ri o",
+ "r io",
+ "▁inter pret",
+ "▁ interpret",
+ "▁slight ly",
+ "fi x",
+ "f ix",
+ "▁stud ies",
+ "▁r id",
+ "▁ri d",
+ "▁ rid",
+ "at re",
+ "atr e",
+ "a tre",
+ "▁benef its",
+ "▁benefit s",
+ "▁F ace",
+ "▁Fa ce",
+ "▁Fac e",
+ "▁ Face",
+ "iv ery",
+ "ive ry",
+ "iver y",
+ "i very",
+ "ри я",
+ "doc ument",
+ "d ocument",
+ "▁as king",
+ "▁ask ing",
+ "La st",
+ "L ast",
+ "ar ante",
+ "ara nte",
+ "aran te",
+ "▁Mart in",
+ "▁E ll",
+ "▁El l",
+ "▁v ector",
+ "▁ve ctor",
+ "▁vec tor",
+ "▁ vector",
+ "▁for ced",
+ "▁force d",
+ "▁ forced",
+ "о ло",
+ "P H",
+ "W R",
+ "▁K l",
+ "▁s ky",
+ "▁sk y",
+ "▁ sky",
+ "▁str ategy",
+ "▁strateg y",
+ "▁strat egy",
+ "oc ked",
+ "ock ed",
+ "▁ne ck",
+ "ś ci",
+ "O UT",
+ ")) ,",
+ ") ),",
+ "C ustom",
+ "▁w ie",
+ "▁ wie",
+ "▁s weet",
+ "▁swe et",
+ "▁t emp",
+ "▁te mp",
+ "▁tem p",
+ "▁ temp",
+ "▁fore ign",
+ "▁h all",
+ "▁ha ll",
+ "▁hal l",
+ "▁ hall",
+ "as tr",
+ "ast r",
+ "a str",
+ "As s",
+ "A ss",
+ "MO DE",
+ "MOD E",
+ "▁max imum",
+ "▁maxim um",
+ "an nels",
+ "ann els",
+ "annel s",
+ "anne ls",
+ "▁t ip",
+ "▁ti p",
+ "▁ tip",
+ "▁second s",
+ "▁sec onds",
+ "▁ seconds",
+ "▁st ack",
+ "▁sta ck",
+ "▁ stack",
+ "ig a",
+ "i ga",
+ "▁r aise",
+ "▁rais e",
+ "▁ra ise",
+ "▁ raise",
+ "en able",
+ "ena ble",
+ "oi r",
+ "o ir",
+ "▁s oul",
+ "▁so ul",
+ "▁sou l",
+ "K e",
+ ")$ .",
+ ") $.",
+ "▁T im",
+ "▁Ti m",
+ "▁ Tim",
+ "AL SE",
+ "is er",
+ "ise r",
+ "i ser",
+ "cont in",
+ "be l",
+ "b el",
+ "▁m ad",
+ "▁ma d",
+ "▁ mad",
+ "lic hen",
+ "li chen",
+ "lich en",
+ "liche n",
+ "l ichen",
+ "ab e",
+ "a be",
+ "sa fe",
+ "▁con cent",
+ "▁conc ent",
+ "▁conce nt",
+ "bo und",
+ "b ound",
+ "▁R equ",
+ "▁Re qu",
+ "▁ Requ",
+ "sw itch",
+ "▁st one",
+ "▁sto ne",
+ "▁ stone",
+ "▁trans l",
+ "▁ transl",
+ "▁v ac",
+ "▁va c",
+ "an don",
+ "and on",
+ "ando n",
+ "▁F ore",
+ "▁For e",
+ "▁Fo re",
+ "▁ Fore",
+ "▁s ounds",
+ "▁sound s",
+ "▁P op",
+ "▁Po p",
+ "▁ Pop",
+ "▁H T",
+ "▁ HT",
+ "li a",
+ "l ia",
+ "en ter",
+ "ent er",
+ "ente r",
+ "▁hel ps",
+ "▁help s",
+ "ed y",
+ "e dy",
+ "ст вен",
+ "ств ен",
+ "стве н",
+ "an ted",
+ "ant ed",
+ "ante d",
+ "▁I ts",
+ "▁It s",
+ "▁St ep",
+ "▁Ste p",
+ "▁ Step",
+ "I con",
+ "▁EX PECT",
+ "▁ EXPECT",
+ "ial ized",
+ "ialize d",
+ "Pos t",
+ "Po st",
+ "P ost",
+ "az e",
+ "a ze",
+ "▁Car ol",
+ "▁Ca rol",
+ "▁re q",
+ "▁r eq",
+ "▁ req",
+ "▁crit ical",
+ "▁critic al",
+ "D S",
+ "▁se at",
+ "▁sea t",
+ "ap ed",
+ "ape d",
+ "a ped",
+ "▁up per",
+ "▁upp er",
+ "▁ upper",
+ "▁S y",
+ "▁ Sy",
+ "▁ex plain",
+ "▁expl ain",
+ "▁' ./",
+ "▁'. /",
+ "ut ils",
+ "util s",
+ "uti ls",
+ "poss ible",
+ "▁d ont",
+ "▁do nt",
+ "▁don t",
+ "H ost",
+ "▁appro xim",
+ "▁approx im",
+ "As ync",
+ "A sync",
+ "▁g rab",
+ "▁gr ab",
+ "▁gra b",
+ "▁s ources",
+ "▁source s",
+ "▁sour ces",
+ "▁ sources",
+ "▁M os",
+ "▁Mo s",
+ "▁Germ any",
+ "▁German y",
+ "▁Ger many",
+ "▁r ub",
+ "▁ru b",
+ "▁ rub",
+ "CH AN",
+ "▁r ain",
+ "▁ra in",
+ "▁tr uly",
+ "▁join ed",
+ "▁jo ined",
+ "▁< ?",
+ "▁ ",
+ "▁L o",
+ "▁ Lo",
+ "Des cription",
+ "De scription",
+ "ak t",
+ "a kt",
+ "▁A nn",
+ "▁An n",
+ "▁ Ann",
+ "^ *",
+ "id ae",
+ "ida e",
+ "( :",
+ "t w",
+ "Ma r",
+ "M ar",
+ "pro du",
+ "prod u",
+ "p rodu",
+ "▁sp oke",
+ "▁spo ke",
+ "ю т",
+ "▁walk ing",
+ "▁wal king",
+ "▁nod ded",
+ "Pro ps",
+ "Pr ops",
+ "Prop s",
+ "En abled",
+ "Enable d",
+ "ir k",
+ "FI LE",
+ "FIL E",
+ "F ILE",
+ "equ al",
+ "eq ual",
+ "e qual",
+ "pp ing",
+ "p ping",
+ "ol i",
+ "o li",
+ "E V",
+ "en z",
+ "et ing",
+ "eti ng",
+ "e ting",
+ "▁s ample",
+ "▁sam ple",
+ "▁ sample",
+ "▁art ist",
+ "[ $",
+ "it à",
+ "й о",
+ "pro ps",
+ "pr ops",
+ "prop s",
+ "b u",
+ "е в",
+ "▁respons ible",
+ "M T",
+ "▁caus ed",
+ "▁cause d",
+ "▁ca used",
+ "▁the me",
+ "▁th eme",
+ "▁them e",
+ "▁ theme",
+ "▁W as",
+ "▁Wa s",
+ "▁ Was",
+ "▁B efore",
+ "▁Be fore",
+ "▁ Before",
+ "ac le",
+ "acl e",
+ "a cle",
+ "▁ро ку",
+ "c u",
+ "DE V",
+ "D EV",
+ "▁h ung",
+ "▁hun g",
+ "▁ hung",
+ "text bf",
+ "▁s pin",
+ "▁sp in",
+ "▁ spin",
+ "▁la test",
+ "▁late st",
+ "▁lat est",
+ "▁ latest",
+ "ent ially",
+ "ential ly",
+ "enti ally",
+ "▁Pro gram",
+ "▁Pr ogram",
+ "▁ Program",
+ "Met adata",
+ "Meta data",
+ "pass word",
+ "▁h urt",
+ "▁hur t",
+ "к с",
+ "▁A us",
+ "▁Au s",
+ "se y",
+ "s ey",
+ "al let",
+ "all et",
+ "alle t",
+ "x F",
+ "▁R oad",
+ "▁Ro ad",
+ "ет ся",
+ "е тся",
+ "▁re nt",
+ "▁r ent",
+ "▁ren t",
+ "▁ rent",
+ "ци я",
+ "▁As sert",
+ "▁Ass ert",
+ "▁ Assert",
+ "і ль",
+ "ü ck",
+ "▁s ites",
+ "▁sit es",
+ "▁si tes",
+ "▁site s",
+ "Doc ument",
+ "D ocument",
+ "▁obt ained",
+ "▁obtain ed",
+ "▁c i",
+ "▁ ci",
+ "▁[ \"",
+ "▁ [\"",
+ "▁com pleted",
+ "▁comp leted",
+ "▁complet ed",
+ "▁compl eted",
+ "▁complete d",
+ "as et",
+ "ase t",
+ "a set",
+ "ra id",
+ "rai d",
+ "r aid",
+ "▁s orry",
+ "▁sor ry",
+ "▁f ab",
+ "▁fa b",
+ "▁ fab",
+ "▁sch ools",
+ "▁school s",
+ "хо ди",
+ "ход и",
+ "▁s cr",
+ "▁sc r",
+ "▁ scr",
+ "▁in cor",
+ "▁inc or",
+ "▁' /",
+ "▁s pr",
+ "▁sp r",
+ "▁ spr",
+ "▁T ext",
+ "▁Te xt",
+ "▁Tex t",
+ "▁ Text",
+ "▁com mercial",
+ "▁commer cial",
+ "in gly",
+ "ing ly",
+ "▁opin ion",
+ "▁S tar",
+ "▁St ar",
+ "▁Sta r",
+ "▁ Star",
+ "Si gn",
+ "Sig n",
+ "S ign",
+ "▁j avax",
+ "▁java x",
+ "▁ javax",
+ "w i",
+ "la t",
+ "l at",
+ "▁K ey",
+ "▁Ke y",
+ "▁ Key",
+ "var phi",
+ "д ы",
+ "▁conne cted",
+ "▁connect ed",
+ "▁ connected",
+ "▁ad just",
+ "▁adj ust",
+ "▁ adjust",
+ "▁A z",
+ "▁ Az",
+ "▁pl anning",
+ "▁plan ning",
+ "-- -",
+ "- --",
+ "In teger",
+ "au f",
+ "a uf",
+ "ex pected",
+ "expect ed",
+ "e xpected",
+ "▁f ant",
+ "▁fa nt",
+ "▁fan t",
+ "▁t ou",
+ "▁to u",
+ "Par ent",
+ "P arent",
+ "▁L at",
+ "▁La t",
+ "▁ Lat",
+ "▁thought s",
+ "▁though ts",
+ "▁J ud",
+ "▁Ju d",
+ "Param eters",
+ "Parameter s",
+ "G r",
+ "ро м",
+ "I A",
+ "▁B ob",
+ "▁Bo b",
+ "lic t",
+ "li ct",
+ "l ict",
+ "la n",
+ "l an",
+ "om ic",
+ "omi c",
+ "o mic",
+ "▁a part",
+ "▁ap art",
+ "▁t rou",
+ "▁tr ou",
+ "▁tro u",
+ "▁app reci",
+ "▁Christ mas",
+ "ir q",
+ "i rq",
+ "th on",
+ "t hon",
+ "▁Er ror",
+ "▁Err or",
+ "▁ Error",
+ "▁s core",
+ "▁sc ore",
+ "▁ score",
+ "ro me",
+ "rom e",
+ "r ome",
+ "▁ne ighbor",
+ "▁neigh bor",
+ "▁neighb or",
+ "▁M ur",
+ "▁Mu r",
+ "ad min",
+ "▁Fil m",
+ "▁Fi lm",
+ "Re ct",
+ "Rec t",
+ "R ect",
+ "▁config uration",
+ "▁ configuration",
+ "▁c s",
+ "▁ cs",
+ "gu n",
+ "g un",
+ "ch annel",
+ "chan nel",
+ "▁Re port",
+ "▁Rep ort",
+ "▁ Report",
+ "▁str ateg",
+ "▁strat eg",
+ "▁work ers",
+ "▁wor kers",
+ "▁worker s",
+ "▁ workers",
+ "field s",
+ "Sch ema",
+ "Sche ma",
+ "S chema",
+ "ap pa",
+ "app a",
+ "ol ic",
+ "oli c",
+ "o lic",
+ "E O",
+ "▁Ch arl",
+ "▁Char l",
+ "▁Cha rl",
+ "▁C up",
+ "▁Cu p",
+ "pn g",
+ "p ng",
+ "▁H ill",
+ "▁Hi ll",
+ "▁Hil l",
+ "ow e",
+ "o we",
+ "▁most ly",
+ "” .",
+ "▁fin ish",
+ "▁ finish",
+ "▁С о",
+ "▁st ars",
+ "▁star s",
+ "▁sta rs",
+ "pl ayer",
+ "play er",
+ "p layer",
+ "▁in ner",
+ "▁inn er",
+ "▁ inner",
+ "com ponent",
+ "ti m",
+ "t im",
+ "I E",
+ "▁t her",
+ "▁the r",
+ "▁th er",
+ "▁ ther",
+ "▁s mart",
+ "▁sm art",
+ "▁ smart",
+ "▁s ad",
+ "▁sa d",
+ "▁Coun cil",
+ "ar ea",
+ "are a",
+ "a rea",
+ "la y",
+ "l ay",
+ "▁б а",
+ "▁ ба",
+ "▁gr adu",
+ "▁grad u",
+ "▁gra du",
+ "▁c hem",
+ "▁ch em",
+ "▁che m",
+ "▁ chem",
+ "▁h o",
+ "▁ ho",
+ "Se lect",
+ "S elect",
+ "▁in str",
+ "▁inst r",
+ "▁ins tr",
+ "▁ instr",
+ "▁k l",
+ "▁ kl",
+ "if ications",
+ "ific ations",
+ "ification s",
+ "Lo ng",
+ "L ong",
+ "▁s obre",
+ "▁so bre",
+ "▁sob re",
+ "▁O ld",
+ "▁Ol d",
+ "▁ Old",
+ "we st",
+ "w est",
+ "}, \\",
+ "} ,\\",
+ "in gu",
+ "ing u",
+ "▁sp ring",
+ "▁spr ing",
+ "▁ spring",
+ "▁n ur",
+ "▁nu r",
+ "ex ample",
+ "Wh en",
+ "Whe n",
+ "W hen",
+ "▁adv ice",
+ "▁u lt",
+ "▁ul t",
+ "▁ ult",
+ "en nis",
+ "enn is",
+ "▁L ove",
+ "▁Lo ve",
+ "▁Lov e",
+ "▁ Love",
+ "▁\" \"",
+ "▁ \"\"",
+ "▁incre ased",
+ "▁increase d",
+ "▁f inding",
+ "▁fin ding",
+ "▁find ing",
+ "ir ty",
+ "irt y",
+ "ist rict",
+ "istr ict",
+ "i strict",
+ "▁l ayer",
+ "▁la yer",
+ "▁lay er",
+ "▁ layer",
+ "temp late",
+ "t emplate",
+ "F irst",
+ "ны м",
+ "igr ation",
+ "ren cy",
+ "r ency",
+ "ow ie",
+ "owi e",
+ "o wie",
+ "▁n p",
+ "▁ np",
+ "▁s election",
+ "▁se lection",
+ "▁select ion",
+ "▁sel ection",
+ "▁sele ction",
+ "▁ selection",
+ "▁N ach",
+ "▁Na ch",
+ "▁P RO",
+ "▁PR O",
+ "▁ PRO",
+ "▁p olic",
+ "▁pol ic",
+ "▁po lic",
+ "▁data base",
+ "▁dat abase",
+ "▁ database",
+ "▁by te",
+ "▁ byte",
+ "▁prov iding",
+ "ma c",
+ "m ac",
+ "▁me tal",
+ "▁met al",
+ "▁meta l",
+ "mod ules",
+ "module s",
+ "▁Ge org",
+ "▁S a",
+ "▁ Sa",
+ "▁est ablish",
+ "▁estab lish",
+ ".. .\"",
+ "... \"",
+ "i u",
+ "ki n",
+ "k in",
+ "▁e th",
+ "▁et h",
+ "▁ eth",
+ "▁S and",
+ "▁San d",
+ "▁Sa nd",
+ "▁Ch apter",
+ "▁Chap ter",
+ "▁g al",
+ "▁ga l",
+ "▁ gal",
+ "▁i ce",
+ "▁ic e",
+ "▁ ice",
+ "Re d",
+ "R ed",
+ "▁d al",
+ "▁da l",
+ "▁ dal",
+ "▁pr incipal",
+ "▁princip al",
+ "Ms g",
+ "M sg",
+ "▁rem ains",
+ "▁remain s",
+ "н г",
+ "T itle",
+ "Re l",
+ "R el",
+ "Dis play",
+ "No n",
+ "N on",
+ "▁def inition",
+ "▁definit ion",
+ "▁defin ition",
+ "▁ definition",
+ "▁at tr",
+ "▁att r",
+ "▁ attr",
+ "▁sign al",
+ "▁sig nal",
+ "▁ signal",
+ "h l",
+ "▁s el",
+ "▁se l",
+ "▁ sel",
+ "▁vol ume",
+ "▁ volume",
+ "▁c ache",
+ "▁ca che",
+ "▁ cache",
+ "he ns",
+ "hen s",
+ "h ens",
+ "▁w ird",
+ "▁wir d",
+ "[ \\",
+ "NO T",
+ "N OT",
+ "▁e lection",
+ "▁el ection",
+ "▁elect ion",
+ "▁ele ction",
+ "▁ election",
+ "ut t",
+ "u tt",
+ "▁W indow",
+ "▁Wind ow",
+ "▁ Window",
+ "en tal",
+ "ent al",
+ "enta l",
+ "if est",
+ "ife st",
+ "x f",
+ "▁Р а",
+ "▁over all",
+ "bl ic",
+ "b lic",
+ "▁ed itor",
+ "▁edit or",
+ "▁ editor",
+ "ad en",
+ "ade n",
+ "a den",
+ "▁c art",
+ "▁car t",
+ "▁ca rt",
+ "▁ cart",
+ "Le ft",
+ "L eft",
+ "ul s",
+ "u ls",
+ "bin g",
+ "bi ng",
+ "b ing",
+ "R ight",
+ "▁s é",
+ "Si m",
+ "S im",
+ "▁came ra",
+ "▁cam era",
+ "▁ camera",
+ "▁f av",
+ "▁fa v",
+ "De cl",
+ "Dec l",
+ "sp ring",
+ "spr ing",
+ "▁err ors",
+ "▁er rors",
+ "▁error s",
+ "▁ errors",
+ "T ab",
+ "print ln",
+ "▁B ern",
+ "▁Be rn",
+ "▁Ber n",
+ "na b",
+ "n ab",
+ "▁B ase",
+ "▁Bas e",
+ "▁Ba se",
+ "▁ Base",
+ "▁a uth",
+ "▁aut h",
+ "▁au th",
+ "▁ auth",
+ "▁app arent",
+ "▁ap parent",
+ "▁appar ent",
+ "▁pres ented",
+ "▁present ed",
+ "▁rem ained",
+ "▁remain ed",
+ "▁w et",
+ "▁we t",
+ "En c",
+ "E nc",
+ "IN FO",
+ "▁S ing",
+ "▁Si ng",
+ "▁Sin g",
+ "▁ Sing",
+ "pack age",
+ ")) );",
+ "))) ;",
+ ") ));",
+ "▁S ocial",
+ "▁So cial",
+ "▁Soc ial",
+ "▁Soci al",
+ "▁M ass",
+ "▁Ma ss",
+ "▁Mas s",
+ "▁ Mass",
+ "▁des pite",
+ "▁desp ite",
+ "▁m obile",
+ "▁mob ile",
+ "▁mobil e",
+ "▁ mobile",
+ "▁l abor",
+ "▁la bor",
+ "▁lab or",
+ "G o",
+ "▁e sp",
+ "▁es p",
+ "▁ esp",
+ "▁T able",
+ "▁Ta ble",
+ "▁Tab le",
+ "▁ Table",
+ "▁ex pert",
+ "▁exper t",
+ "▁exp ert",
+ "▁f lex",
+ "▁fl ex",
+ "▁fle x",
+ "▁ flex",
+ "▁prof ession",
+ "▁profess ion",
+ "▁p il",
+ "▁pi l",
+ "Col lection",
+ "Coll ection",
+ "Collect ion",
+ "LO CK",
+ "LOC K",
+ "▁ap plied",
+ "▁appl ied",
+ "al ler",
+ "all er",
+ "alle r",
+ "or ph",
+ "orp h",
+ "EN SE",
+ "ENS E",
+ "▁бы л",
+ "▁d b",
+ "▁ db",
+ "over line",
+ "▁C ode",
+ "▁Co de",
+ "▁ Code",
+ "▁by tes",
+ "▁byte s",
+ "▁ bytes",
+ "▁tr ouble",
+ "▁trou ble",
+ "▁на се",
+ "D D",
+ "▁Y ear",
+ "▁Ye ar",
+ "▁ Year",
+ "mb ox",
+ "m box",
+ "▁ke eping",
+ "▁keep ing",
+ "▁ keeping",
+ "▁k ick",
+ "▁ki ck",
+ "än g",
+ "ä ng",
+ "▁correspon ding",
+ "▁correspond ing",
+ "▁l ibrary",
+ "▁ library",
+ "▁*/ \r",
+ "▁ */\r",
+ "call back",
+ "um s",
+ "u ms",
+ "▁j son",
+ "▁js on",
+ "▁ json",
+ "▁M ount",
+ "▁Mo unt",
+ "▁ Mount",
+ "▁St and",
+ "▁Stan d",
+ "▁Sta nd",
+ "▁ Stand",
+ "IG HT",
+ "IGH T",
+ "▁New s",
+ "▁Ne ws",
+ "▁ News",
+ "▁com ments",
+ "▁comm ents",
+ "▁comment s",
+ "▁ comments",
+ "return s",
+ "C al",
+ "▁a ward",
+ "▁aw ard",
+ "▁b ought",
+ "▁bou ght",
+ "include graphics",
+ "▁ ле",
+ "do t",
+ "d ot",
+ "ro nic",
+ "ron ic",
+ "r onic",
+ "▁extrem ely",
+ "▁extreme ly",
+ "▁min or",
+ "▁mi nor",
+ "if er",
+ "ife r",
+ "i fer",
+ "ja va",
+ "jav a",
+ "j ava",
+ "en dar",
+ "end ar",
+ "enda r",
+ "la yout",
+ "lay out",
+ "l ayout",
+ "pl ies",
+ "▁b uf",
+ "▁bu f",
+ "▁ buf",
+ "▁Is land",
+ "▁Ab out",
+ "▁ About",
+ "▁w est",
+ "▁we st",
+ "▁ west",
+ "▁S cott",
+ "▁Sc ott",
+ "▁Scot t",
+ "AC T",
+ "A CT",
+ "Wh y",
+ "W hy",
+ "▁large st",
+ "▁larg est",
+ "▁cont ainer",
+ "▁contain er",
+ "▁ container",
+ "▁t emperature",
+ "▁temper ature",
+ "▁ £",
+ "▁red uce",
+ "▁redu ce",
+ "▁ reduce",
+ "▁f oi",
+ "▁fo i",
+ "ha n",
+ "h an",
+ "▁b od",
+ "▁bo d",
+ "▁V an",
+ "▁Va n",
+ "▁null ptr",
+ "▁ nullptr",
+ "▁d ating",
+ "▁da ting",
+ "▁dat ing",
+ "▁ dating",
+ "▁ch ain",
+ "▁cha in",
+ "▁ chain",
+ "Fl ags",
+ "Flag s",
+ "ient o",
+ "ien to",
+ "i ento",
+ "so rt",
+ "sor t",
+ "s ort",
+ "▁f an",
+ "▁fa n",
+ "▁ fan",
+ "▁det ermine",
+ "▁determ ine",
+ "▁determin e",
+ "▁deter mine",
+ "▁w ear",
+ "▁we ar",
+ "▁ wear",
+ "B E",
+ "▁appropri ate",
+ "л ся",
+ "то в",
+ "т ов",
+ "▁go als",
+ "▁goal s",
+ "▁M ap",
+ "▁Ma p",
+ "▁ Map",
+ "▁S ar",
+ "▁Sa r",
+ "▁O ption",
+ "▁Opt ion",
+ "▁ Option",
+ "▁h ate",
+ "▁ha te",
+ "▁hat e",
+ "▁z ijn",
+ ", -",
+ "▁im plied",
+ "▁impl ied",
+ "bit s",
+ "bi ts",
+ "b its",
+ "▁M en",
+ "▁Me n",
+ "▁ Men",
+ "sk ip",
+ "ski p",
+ "▁M ond",
+ "▁Mon d",
+ "▁Mo nd",
+ "▁H on",
+ "▁Ho n",
+ "▁pro ve",
+ "▁pr ove",
+ "▁prov e",
+ "va n",
+ "v an",
+ "▁tr aff",
+ "▁tra ff",
+ "▁in tr",
+ "▁int r",
+ "▁ intr",
+ "pi c",
+ "p ic",
+ "▁dro pped",
+ "▁drop ped",
+ "▁w erd",
+ "▁we rd",
+ "▁wer d",
+ "▁separ ate",
+ "is a",
+ "i sa",
+ "▁t ab",
+ "▁ta b",
+ "▁ tab",
+ "tm l",
+ "t ml",
+ "▁\" $",
+ "mu tex",
+ "mut ex",
+ "▁P an",
+ "▁Pa n",
+ "▁ Pan",
+ "ser ve",
+ "serv e",
+ "s erve",
+ "▁hot el",
+ "▁L ast",
+ "▁La st",
+ "▁Las t",
+ "▁ Last",
+ "st ep",
+ "ste p",
+ "▁v ir",
+ "▁vi r",
+ "▁ vir",
+ "R ule",
+ "is tan",
+ "ist an",
+ "ista n",
+ "i stan",
+ "ot ing",
+ "oti ng",
+ "o ting",
+ "ar ks",
+ "ark s",
+ "(_ _",
+ "( __",
+ "▁e ls",
+ "▁el s",
+ "▁ els",
+ "Pl ayer",
+ "Play er",
+ "P layer",
+ "] ]",
+ "ви ч",
+ "yc h",
+ "y ch",
+ "ex ception",
+ "except ion",
+ "=\" ../",
+ "▁im agine",
+ "▁imag ine",
+ "▁imagin e",
+ "\"} ,",
+ "\" },",
+ "ic ago",
+ "ica go",
+ "el er",
+ "ele r",
+ "e ler",
+ "▁v s",
+ "▁ vs",
+ "▁A frica",
+ "▁Afr ica",
+ "▁Bus iness",
+ "oc ks",
+ "ock s",
+ "o cks",
+ "▁p rz",
+ "▁pr z",
+ "▁fuck ing",
+ "▁p icked",
+ "▁pick ed",
+ "▁pic ked",
+ "▁в і",
+ "▁ ві",
+ "▁\" ,",
+ "▁ \",",
+ "▁b ott",
+ "▁bo tt",
+ "▁bot t",
+ "▁fail ure",
+ "▁ failure",
+ "[ :",
+ "▁G ar",
+ "▁Ga r",
+ "ap es",
+ "ape s",
+ "a pes",
+ "up le",
+ "u ple",
+ "▁f er",
+ "▁fe r",
+ "▁ fer",
+ "▁p urchase",
+ "▁purch ase",
+ "▁п ер",
+ "▁пе р",
+ "▁ пер",
+ "▁b ird",
+ "▁bi rd",
+ "▁ bird",
+ "W idget",
+ "▁Sund ay",
+ "▁Sun day",
+ "▁A maz",
+ "▁Am az",
+ "▁ Amaz",
+ "▁cons ult",
+ "ut sch",
+ "uts ch",
+ "an to",
+ "ant o",
+ "St orage",
+ "▁he ader",
+ "▁head er",
+ "▁ header",
+ "üh r",
+ "ü hr",
+ "▁H a",
+ "▁ Ha",
+ "▁Associ ation",
+ "▁s ight",
+ "▁si ght",
+ "▁sig ht",
+ "▁sigh t",
+ "C ell",
+ "▁pro file",
+ "▁prof ile",
+ "▁ profile",
+ "▁fem ale",
+ "å n",
+ "▁w id",
+ "▁ wid",
+ "z n",
+ "Dir ect",
+ "Di rect",
+ "D irect",
+ "▁st ret",
+ "▁str et",
+ "▁stre t",
+ "▁ stret",
+ "aa t",
+ "a at",
+ "▁pat ient",
+ "▁ patient",
+ "he re",
+ "her e",
+ "h ere",
+ "▁A tl",
+ "▁At l",
+ "in et",
+ "ine t",
+ "i net",
+ "Def inition",
+ "im ary",
+ "ima ry",
+ "i mary",
+ "Pol icy",
+ "▁d ut",
+ "▁du t",
+ "▁major ity",
+ "с і",
+ "▁Pro ject",
+ "▁ Project",
+ "By Id",
+ "▁belie ved",
+ "▁believe d",
+ "▁Mus ic",
+ "▁ Music",
+ "з ы",
+ "an ti",
+ "ant i",
+ "▁o der",
+ "▁od er",
+ "▁ oder",
+ "Ch annel",
+ "▁s le",
+ "▁sl e",
+ "▁sequ ence",
+ "▁ sequence",
+ "▁pie ces",
+ "▁piece s",
+ "▁k ne",
+ "▁kn e",
+ "▁abs olutely",
+ "▁absolut ely",
+ "▁absolute ly",
+ "▁Phil ip",
+ "ab ilities",
+ "abil ities",
+ "Qu e",
+ "Q ue",
+ "▁K ar",
+ "▁Ka r",
+ "Ex ecut",
+ "Exec ut",
+ "▁D evel",
+ "▁De vel",
+ "▁Dev el",
+ "▁elect ric",
+ "ful l",
+ "fu ll",
+ "f ull",
+ "rol led",
+ "roll ed",
+ "Do m",
+ "D om",
+ "▁r iver",
+ "▁ri ver",
+ "▁riv er",
+ "▁ river",
+ "▁health y",
+ "▁heal thy",
+ "▁ex tern",
+ "▁ext ern",
+ "fi t",
+ "f it",
+ "▁co ach",
+ "▁K r",
+ "as ta",
+ "ast a",
+ "a sta",
+ "Com pat",
+ "Comp at",
+ "▁e xit",
+ "▁ex it",
+ "▁ exit",
+ "▁Con st",
+ "▁Cons t",
+ "▁ Const",
+ "af ter",
+ "aft er",
+ "a fter",
+ "▁should er",
+ "▁j obs",
+ "▁job s",
+ "▁jo bs",
+ "zo ne",
+ "zon e",
+ "z one",
+ "▁s ale",
+ "▁sa le",
+ "▁sal e",
+ "ix el",
+ "▁determ ined",
+ "▁determine d",
+ "▁determin ed",
+ "▁any way",
+ "or f",
+ "o rf",
+ "▁G er",
+ "▁Ge r",
+ "all el",
+ "alle l",
+ "re es",
+ "ree s",
+ "r ees",
+ "as m",
+ "a sm",
+ "im s",
+ "i ms",
+ "▁rec ords",
+ "▁record s",
+ "▁ records",
+ "▁cor por",
+ "▁int ellig",
+ "▁intel lig",
+ "▁P rem",
+ "▁Pr em",
+ "▁Pre m",
+ "▁d riving",
+ "▁dr iving",
+ "▁dri ving",
+ "▁driv ing",
+ "▁mar riage",
+ "▁Th ank",
+ "▁ Thank",
+ "▁w illing",
+ "▁will ing",
+ "M C",
+ "Field s",
+ "It ems",
+ "Item s",
+ "▁m icro",
+ "▁mi cro",
+ "▁mic ro",
+ "▁l ift",
+ "▁li ft",
+ "▁lif t",
+ "ir ection",
+ "ire ction",
+ "irect ion",
+ "i rection",
+ "Acc ount",
+ "Ac count",
+ "▁arch itect",
+ "tr ack",
+ "tra ck",
+ "▁p rin",
+ "▁pr in",
+ "▁pri n",
+ "P A",
+ "▁r uns",
+ "▁run s",
+ "▁ru ns",
+ "▁Tex as",
+ "is her",
+ "ish er",
+ "en sure",
+ "ens ure",
+ "▁B oth",
+ "▁Bo th",
+ "▁Bot h",
+ "ко м",
+ "▁Col or",
+ "▁Co lor",
+ "▁ Color",
+ "Reg ister",
+ "▁J oe",
+ "▁Jo e",
+ "ge q",
+ "g eq",
+ "le ts",
+ "let s",
+ "l ets",
+ "ad ing",
+ "adi ng",
+ "a ding",
+ "▁ar my",
+ "▁arm y",
+ "▁B ank",
+ "▁Ban k",
+ "▁ Bank",
+ "ot ic",
+ "oti c",
+ "Pro duct",
+ "Produ ct",
+ "im port",
+ "imp ort",
+ "▁W ed",
+ "▁We d",
+ "▁c ry",
+ "▁cr y",
+ "gr ade",
+ "grad e",
+ "gra de",
+ "g rade",
+ "di g",
+ "d ig",
+ "ga l",
+ "g al",
+ "к ла",
+ "es ted",
+ "est ed",
+ "este d",
+ "e sted",
+ "õ es",
+ "ge rs",
+ "ger s",
+ "g ers",
+ "olog ie",
+ "olo gie",
+ "то м",
+ "ra zy",
+ "raz y",
+ "r azy",
+ "▁d inner",
+ "▁din ner",
+ "Q U",
+ "▁fin gers",
+ "▁fing ers",
+ "▁finger s",
+ "UL E",
+ "U LE",
+ "cl aim",
+ "▁adv antage",
+ "▁advant age",
+ "▁var iable",
+ "▁vari able",
+ "▁ variable",
+ "▁med ic",
+ "▁medi c",
+ "▁m ale",
+ "▁ma le",
+ "▁mal e",
+ "▁circ um",
+ "▁м і",
+ "▁ мі",
+ "▁inter net",
+ "▁intern et",
+ "W N",
+ "▁l ab",
+ "▁la b",
+ "▁ lab",
+ "az ine",
+ "azi ne",
+ "ч но",
+ "▁l oop",
+ "▁lo op",
+ "▁ loop",
+ "▁p red",
+ "▁pre d",
+ "▁pr ed",
+ "▁ pred",
+ "▁con sequ",
+ "▁cons equ",
+ "▁conse qu",
+ "▁bal ance",
+ "▁ balance",
+ "fort un",
+ "▁g ift",
+ "▁gi ft",
+ "▁d rug",
+ "▁dr ug",
+ "▁dru g",
+ "▁c ash",
+ "▁cas h",
+ "▁ca sh",
+ "ски х",
+ "с ких",
+ "r g",
+ "ist ribut",
+ "▁high est",
+ "▁hig hest",
+ "êm e",
+ "ê me",
+ "em ph",
+ "emp h",
+ "em on",
+ "e mon",
+ "▁per formed",
+ "▁perform ed",
+ "cu t",
+ "c ut",
+ "▁cl oser",
+ "▁close r",
+ "▁clos er",
+ "▁clo ser",
+ "▁be coming",
+ "▁bec oming",
+ "▁\" \",",
+ "▁\"\" ,",
+ "st ar",
+ "sta r",
+ "s tar",
+ "pu b",
+ "p ub",
+ "▁pre par",
+ "▁prep ar",
+ "▁v ote",
+ "▁vo te",
+ "▁vot e",
+ "▁ vote",
+ "il de",
+ "ild e",
+ "▁im press",
+ "▁imp ress",
+ "▁employ ees",
+ "▁employee s",
+ "▁e inen",
+ "▁ein en",
+ "▁eine n",
+ "▁sm ooth",
+ "▁s now",
+ "▁sn ow",
+ "▁p urs",
+ "▁pur s",
+ "▁pu rs",
+ "▁v oc",
+ "▁vo c",
+ "▁M icrosoft",
+ "▁Micro soft",
+ "▁ Microsoft",
+ "P U",
+ "▁in come",
+ "▁inc ome",
+ "in os",
+ "ino s",
+ "i nos",
+ "▁oper ator",
+ "▁opera tor",
+ "▁ operator",
+ "▁equ ival",
+ "▁pass word",
+ "▁ password",
+ "ci ón",
+ "ció n",
+ "c ión",
+ "su ccess",
+ "▁e mp",
+ "▁em p",
+ "▁ emp",
+ "HO UT",
+ "H OUT",
+ "▁c a",
+ "▁ ca",
+ "fl ag",
+ "f lag",
+ "il ly",
+ "ill y",
+ "cre te",
+ "cr ete",
+ "cret e",
+ "fr ak",
+ "▁h idden",
+ "▁hid den",
+ "▁ hidden",
+ "▁\" %",
+ "▁ \"%",
+ "ER N",
+ "ро ва",
+ "ров а",
+ "▁U N",
+ "▁ UN",
+ "ro ke",
+ "rok e",
+ "r oke",
+ "mi ss",
+ "m iss",
+ "▁s plit",
+ "▁sp lit",
+ "▁spl it",
+ "▁ split",
+ "Re ference",
+ ")$ ,",
+ ") $,",
+ "ep er",
+ "e per",
+ "▁N O",
+ "▁ NO",
+ "▁s quare",
+ "▁squ are",
+ "▁ square",
+ "su r",
+ "s ur",
+ "че н",
+ "ч ен",
+ "es ter",
+ "est er",
+ "este r",
+ "e ster",
+ "н ь",
+ "} \"",
+ "ra wn",
+ "raw n",
+ "r awn",
+ "ru le",
+ "r ule",
+ "▁aud ience",
+ "es te",
+ "est e",
+ "e ste",
+ "em s",
+ "e ms",
+ "IC ENSE",
+ "▁I ll",
+ "▁Il l",
+ "▁ Ill",
+ "US E",
+ "U SE",
+ "▁b on",
+ "▁bo n",
+ "▁ bon",
+ "bu r",
+ "b ur",
+ "▁s ick",
+ "▁si ck",
+ "▁h orse",
+ "▁hor se",
+ "▁hors e",
+ "▁E duc",
+ "▁Ed uc",
+ "▁Edu c",
+ "▁benef it",
+ "▁c ro",
+ "▁cr o",
+ "▁ cro",
+ "Ap plication",
+ "▁cor re",
+ "▁gu arante",
+ "DA TA",
+ "DAT A",
+ "D ATA",
+ "▁expl ained",
+ "▁explain ed",
+ "T X",
+ "▁o nt",
+ "▁on t",
+ "▁ ont",
+ "▁F lor",
+ "▁Fl or",
+ "▁Flo r",
+ "▁re ports",
+ "▁rep orts",
+ "▁report s",
+ "▁Re al",
+ "▁ Real",
+ "ud ed",
+ "ude d",
+ "u ded",
+ "le an",
+ "▁cit iz",
+ "▁dec ide",
+ "▁decid e",
+ "W S",
+ "▁do main",
+ "▁dom ain",
+ "▁ domain",
+ "▁ref lect",
+ "▁ reflect",
+ "▁min imum",
+ "▁minim um",
+ "▁le gs",
+ "▁leg s",
+ "▁sm iled",
+ "▁smile d",
+ "f i",
+ "▁p ure",
+ "▁pur e",
+ "▁pu re",
+ "▁C ustom",
+ "▁ Custom",
+ "▁ess ential",
+ "▁observ ed",
+ "▁observe d",
+ "▁obs erved",
+ "By tes",
+ "Byte s",
+ "▁c tx",
+ "▁ ctx",
+ "▁r ates",
+ "▁rate s",
+ "▁rat es",
+ "▁ra tes",
+ "mb re",
+ "m bre",
+ "▁w orry",
+ "▁wor ry",
+ ") ^",
+ "▁Re search",
+ "▁Res earch",
+ "Ro ot",
+ "R oot",
+ "Window s",
+ "ult ure",
+ "ultur e",
+ "▁rel ative",
+ "▁relativ e",
+ "▁ relative",
+ "▁s eu",
+ "▁se u",
+ "▁n ie",
+ "▁ni e",
+ "▁ nie",
+ "▁s hook",
+ "▁sh ook",
+ "ious ly",
+ "i ously",
+ "▁ad vert",
+ "▁adv ert",
+ "Se e",
+ "S ee",
+ "▁Cent ral",
+ "▁b atter",
+ "▁batt er",
+ "▁bat ter",
+ "▁s igned",
+ "▁sign ed",
+ "▁sig ned",
+ "▁ signed",
+ "T S",
+ "on i",
+ "o ni",
+ "▁pre pared",
+ "▁prep ared",
+ "▁prepar ed",
+ "▁prepare d",
+ "ga te",
+ "g ate",
+ "▁C are",
+ "▁Car e",
+ "▁Ca re",
+ "ca re",
+ "car e",
+ "c are",
+ "▁sup ply",
+ "▁supp ly",
+ "Ex p",
+ "E xp",
+ "bol ds",
+ "bold s",
+ "b olds",
+ "▁tr ail",
+ "▁tra il",
+ "▁f ish",
+ "▁fi sh",
+ "▁fis h",
+ "▁ fish",
+ "▁un its",
+ "▁unit s",
+ "▁ units",
+ "ven ue",
+ "v enue",
+ "х и",
+ "▁W ood",
+ "▁Wo od",
+ "▁c ategory",
+ "▁categ ory",
+ "▁categor y",
+ "▁ category",
+ "▁b le",
+ "▁bl e",
+ "▁ ble",
+ "▁over ride",
+ "▁ override",
+ "fo o",
+ "f oo",
+ "▁influ ence",
+ "en th",
+ "ent h",
+ "ri j",
+ "r ij",
+ "▁ad apt",
+ "ic ians",
+ "ici ans",
+ "ician s",
+ "icia ns",
+ "de leted",
+ "del eted",
+ "delete d",
+ "▁v ision",
+ "▁vis ion",
+ "▁ vision",
+ "ct rl",
+ "ctr l",
+ "c trl",
+ "L ambda",
+ "t p",
+ "mon d",
+ "mo nd",
+ "m ond",
+ "atur day",
+ "norm al",
+ "nor mal",
+ "n ormal",
+ "▁thous and",
+ "▁Prof ess",
+ "▁dise ase",
+ "cl ip",
+ "cli p",
+ "▁г ра",
+ "▁ гра",
+ "bolds ymbol",
+ "bold symbol",
+ "O B",
+ "▁chall enge",
+ "▁challeng e",
+ "▁m otion",
+ "▁mot ion",
+ "▁w his",
+ "▁wh is",
+ "▁le aders",
+ "▁lead ers",
+ "▁leader s",
+ "▁col on",
+ "▁co lon",
+ "▁ colon",
+ "▁s uit",
+ "▁su it",
+ "▁ suit",
+ "mi d",
+ "m id",
+ "amp ion",
+ "á g",
+ "▁view s",
+ "▁vie ws",
+ "▁ views",
+ "▁app ears",
+ "▁appe ars",
+ "▁appear s",
+ "an cel",
+ "ance l",
+ "anc el",
+ "▁z we",
+ "▁zw e",
+ "IS T",
+ "I ST",
+ "▁le aves",
+ "▁leave s",
+ "▁e nh",
+ "▁en h",
+ "▁ enh",
+ "Act ive",
+ "Activ e",
+ "▁d it",
+ "▁di t",
+ "▁ dit",
+ "if icate",
+ "ific ate",
+ "ifica te",
+ "mat rix",
+ "Ex pression",
+ "Exp ression",
+ "Expr ession",
+ "Express ion",
+ "Re ader",
+ "Read er",
+ "▁m ental",
+ "▁men tal",
+ "▁ment al",
+ "em bre",
+ "emb re",
+ "e mbre",
+ "▁de cor",
+ "▁dec or",
+ "▁ decor",
+ "ar ts",
+ "art s",
+ "▁v ent",
+ "▁ve nt",
+ "▁ven t",
+ "▁ vent",
+ "ne l",
+ "n el",
+ "line s",
+ "li nes",
+ "lin es",
+ "l ines",
+ "up id",
+ "u pid",
+ "er ved",
+ "erv ed",
+ "erve d",
+ "▁bo ys",
+ "▁boy s",
+ "▁ boys",
+ "ал ь",
+ "а ль",
+ "MO D",
+ "M OD",
+ "is l",
+ "i sl",
+ "▁[ [",
+ "▁ [[",
+ "ph y",
+ "p hy",
+ "▁. .",
+ "▁ ..",
+ "▁a gent",
+ "▁ag ent",
+ "▁age nt",
+ "▁ agent",
+ "▁S ervices",
+ "▁Service s",
+ "▁Serv ices",
+ "▁ Services",
+ "▁i ron",
+ "▁ir on",
+ "▁ iron",
+ "▁com ponents",
+ "▁compon ents",
+ "▁component s",
+ "▁ components",
+ "▁f re",
+ "▁fr e",
+ "▁ fre",
+ "iction ary",
+ "▁t ests",
+ "▁te sts",
+ "▁test s",
+ "▁ tests",
+ ".~ \\",
+ ". ~\\",
+ "ob s",
+ "o bs",
+ "▁М и",
+ "▁об ла",
+ "▁ass ess",
+ "▁Fr iday",
+ "▁we ather",
+ "k g",
+ "ст ра",
+ "с тра",
+ ". }",
+ "end ant",
+ "enda nt",
+ "an na",
+ "ann a",
+ "▁Japan ese",
+ "cm p",
+ "c mp",
+ "▁Ar my",
+ "▁Arm y",
+ "on ym",
+ "ony m",
+ "o nym",
+ "▁rel ax",
+ "date s",
+ "da tes",
+ "dat es",
+ "d ates",
+ "▁R ussian",
+ "▁Russ ian",
+ "▁Russia n",
+ "▁excell ent",
+ "') )",
+ "' ))",
+ "IL ITY",
+ "▁sh owing",
+ "▁show ing",
+ "▁Dan iel",
+ "м я",
+ "▁M ain",
+ "▁Ma in",
+ "▁Mai n",
+ "▁ Main",
+ "Ph i",
+ "P hi",
+ "▁R ock",
+ "▁Ro ck",
+ "▁Roc k",
+ "▁g rew",
+ "▁gr ew",
+ "▁gre w",
+ "▁y ield",
+ "i ère",
+ "se g",
+ "s eg",
+ "}} $",
+ "} }$",
+ "▁st rict",
+ "▁str ict",
+ "▁stri ct",
+ "▁ strict",
+ "▁v ehicle",
+ "▁veh icle",
+ "U D",
+ "A F",
+ "S w",
+ "▁c hest",
+ "▁ch est",
+ "▁che st",
+ "▁off icer",
+ "▁offic er",
+ "▁office r",
+ "▁e ar",
+ "▁ ear",
+ "HE R",
+ "H ER",
+ "no on",
+ "n oon",
+ "▁jour ney",
+ "N T",
+ "▁d ivers",
+ "▁di vers",
+ "▁div ers",
+ "▁diver s",
+ "▁dive rs",
+ "▁Fin ally",
+ "▁Final ly",
+ "F ound",
+ "▁A S",
+ "▁ AS",
+ "ri k",
+ "r ik",
+ "▁con str",
+ "▁const r",
+ "▁cons tr",
+ "▁s ust",
+ "▁su st",
+ "▁sus t",
+ "ac count",
+ "acc ount",
+ "acco unt",
+ "▁w alls",
+ "▁wall s",
+ "▁wal ls",
+ "▁entire ly",
+ "It er",
+ "I ter",
+ "ch a",
+ "c ha",
+ "is hes",
+ "ish es",
+ "IV E",
+ "I VE",
+ "▁pr ime",
+ "▁prim e",
+ "▁pri me",
+ "▁ prime",
+ "▁ …",
+ "x e",
+ "ut en",
+ "ute n",
+ "u ten",
+ "ar se",
+ "ars e",
+ "▁P a",
+ "put e",
+ "pu te",
+ "p ute",
+ "ä l",
+ "▁prote ction",
+ "▁protect ion",
+ "▁prot ection",
+ "▁ke ys",
+ "▁key s",
+ "▁ keys",
+ "Ma y",
+ "M ay",
+ "By te",
+ "Con st",
+ "Cons t",
+ "B L",
+ "▁п е",
+ "▁ пе",
+ "▁s pl",
+ "▁sp l",
+ "▁ spl",
+ "▁cl othes",
+ "▁cloth es",
+ "as hed",
+ "ash ed",
+ "Mar k",
+ "M ark",
+ "è me",
+ "▁f ait",
+ "▁fa it",
+ "▁introdu ced",
+ "▁introduce d",
+ "un lock",
+ "▁In stead",
+ "▁Inst ead",
+ "ans ion",
+ "reg ion",
+ "▁Amer icans",
+ "▁American s",
+ "▁America ns",
+ "▁ind eed",
+ "▁inde ed",
+ "wid get",
+ "w idget",
+ "▁real ize",
+ "▁realiz e",
+ "▁f ro",
+ "▁fr o",
+ "BI T",
+ "B IT",
+ "▁Re act",
+ "▁ React",
+ "RE AD",
+ "as ket",
+ "ask et",
+ "ne ver",
+ "n ever",
+ "▁p oll",
+ "▁pol l",
+ "▁po ll",
+ "▁ poll",
+ "ic ol",
+ "ico l",
+ "i col",
+ "▁p rev",
+ "▁pre v",
+ "▁pr ev",
+ "▁ prev",
+ "▁h yp",
+ "▁hy p",
+ "▁F ur",
+ "▁Fu r",
+ "cl oud",
+ "▁L ee",
+ "▁Le e",
+ "pl ing",
+ "p ling",
+ "▁Ch ild",
+ "▁Chi ld",
+ "▁ Child",
+ "▁ide al",
+ "▁idea l",
+ "Se lector",
+ "Select or",
+ "STAT US",
+ "uct ure",
+ "▁w ine",
+ "▁win e",
+ "▁poss ibly",
+ "▁put ting",
+ "▁r iv",
+ "▁ri v",
+ "▁ riv",
+ "▁w earing",
+ "▁we aring",
+ "▁wear ing",
+ "▁S ource",
+ "▁ Source",
+ "▁C as",
+ "▁Ca s",
+ "Ch anged",
+ "Change d",
+ "▁th anks",
+ "▁than ks",
+ "▁thank s",
+ "TI ME",
+ "TIM E",
+ "T IME",
+ "▁s port",
+ "▁sp ort",
+ "▁spo rt",
+ "▁A ward",
+ "▁Aw ard",
+ "▁g lad",
+ "▁gl ad",
+ "▁P ass",
+ "▁Pa ss",
+ "▁Pas s",
+ "▁ Pass",
+ "▁P os",
+ "▁Po s",
+ "▁ Pos",
+ "sc he",
+ "sch e",
+ "s che",
+ "▁C D",
+ "▁ CD",
+ "▁aff ord",
+ "▁af ford",
+ "▁W omen",
+ "▁Wo men",
+ "▁D istrict",
+ "▁Di strict",
+ "▁Dist rict",
+ "▁id entity",
+ "▁ident ity",
+ "▁ identity",
+ "▁part ies",
+ "▁par ties",
+ "▁partie s",
+ "▁parti es",
+ ": %",
+ "▁d rag",
+ "▁dr ag",
+ "▁ drag",
+ "▁m ai",
+ "▁ma i",
+ "! (",
+ "lang le",
+ "lan gle",
+ "l angle",
+ "▁kn owing",
+ "▁know ing",
+ "Pro ject",
+ "▁reg arding",
+ "▁regard ing",
+ "▁Jose ph",
+ "▁Jos eph",
+ "г е",
+ "▁D ar",
+ "▁Da r",
+ "▁H or",
+ "▁Ho r",
+ "▁ Hor",
+ "▁anim als",
+ "▁animal s",
+ "▁ext ension",
+ "▁extens ion",
+ "▁ extension",
+ "ска я",
+ "▁H an",
+ "▁Ha n",
+ "bt n",
+ "b tn",
+ "ac iones",
+ "aci ones",
+ "acion es",
+ "acio nes",
+ "▁f amiliar",
+ "▁fam iliar",
+ "▁famil iar",
+ "▁familia r",
+ "hol der",
+ "hold er",
+ "h older",
+ ": \r",
+ "st ood",
+ "sto od",
+ "▁li ked",
+ "▁like d",
+ "▁lik ed",
+ "CO DE",
+ "▁en able",
+ "▁ enable",
+ "▁p ed",
+ "▁pe d",
+ "▁ ped",
+ "it i",
+ "i ti",
+ "ha b",
+ "h ab",
+ "DI R",
+ "D IR",
+ "▁be at",
+ "▁ beat",
+ "т і",
+ "▁Min ister",
+ "▁Mini ster",
+ "▁p y",
+ "▁ py",
+ "P at",
+ "▁ex hib",
+ "▁exh ib",
+ "▁B uild",
+ "▁Bu ild",
+ "▁ Build",
+ "▁F ield",
+ "▁Fi eld",
+ "▁ Field",
+ "ic ian",
+ "ici an",
+ "icia n",
+ "▁coll abor",
+ "▁qu arter",
+ "▁quart er",
+ "▁quar ter",
+ "▁F alse",
+ "▁Fal se",
+ "▁ False",
+ "k m",
+ "▁v irtual",
+ "▁virt ual",
+ "▁ virtual",
+ "ow a",
+ "o wa",
+ "▁J on",
+ "▁Jo n",
+ "am in",
+ "ami n",
+ "a min",
+ "ue n",
+ "u en",
+ "▁и н",
+ "▁ ин",
+ "im ation",
+ "imat ion",
+ "ov ing",
+ "ovi ng",
+ "o ving",
+ "▁test ing",
+ "▁ testing",
+ "se ct",
+ "sec t",
+ "s ect",
+ "IT ION",
+ "I TION",
+ "! \\",
+ "ap y",
+ "a py",
+ "▁trans ition",
+ "▁transit ion",
+ "▁ transition",
+ "os itory",
+ "OD O",
+ "O DO",
+ "P D",
+ "n é",
+ "▁gener ate",
+ "▁gene rate",
+ "▁ generate",
+ "▁n ative",
+ "▁nat ive",
+ "▁ native",
+ "▁( '",
+ "▁ ('",
+ "▁e lle",
+ "▁el le",
+ "▁ell e",
+ "▁ elle",
+ "R R",
+ "▁h un",
+ "_- >",
+ "_ ->",
+ "ag nost",
+ "agn ost",
+ "▁pro posed",
+ "▁prop osed",
+ "▁propos ed",
+ "▁propose d",
+ "▁G ame",
+ "▁Ga me",
+ "▁Gam e",
+ "▁ Game",
+ "▁eff orts",
+ "▁effort s",
+ "в я",
+ "t c",
+ "с к",
+ "▁int ent",
+ "▁inte nt",
+ "▁ intent",
+ "▁B re",
+ "▁Br e",
+ "is c",
+ "i sc",
+ "▁pro test",
+ "▁prote st",
+ "▁prot est",
+ "▁h olds",
+ "▁hold s",
+ "▁hol ds",
+ "▁ holds",
+ "om etry",
+ "ome try",
+ "omet ry",
+ "o metry",
+ "▁H ave",
+ "▁Ha ve",
+ "▁Hav e",
+ "▁ Have",
+ "▁de tail",
+ "▁det ail",
+ "▁ detail",
+ "▁WIT HOUT",
+ "▁WITH OUT",
+ "ye r",
+ "y er",
+ "▁K on",
+ "▁Ko n",
+ "▁not iced",
+ "▁notice d",
+ "▁require ments",
+ "▁requirement s",
+ "DE BUG",
+ "ki ns",
+ "kin s",
+ "k ins",
+ "▁S pan",
+ "▁Sp an",
+ "▁ Span",
+ "▁c ars",
+ "▁car s",
+ "▁ca rs",
+ "me ta",
+ "met a",
+ "m eta",
+ "▁k il",
+ "▁ki l",
+ "▁ kil",
+ "▁B ron",
+ "▁Br on",
+ "▁Bro n",
+ "▁experience d",
+ "▁experi enced",
+ "▁re mind",
+ "▁rem ind",
+ "our se",
+ "ours e",
+ "▁W estern",
+ "▁West ern",
+ "▁Wes tern",
+ "ter ed",
+ "te red",
+ "tere d",
+ "t ered",
+ "▁dev ices",
+ "▁device s",
+ "▁ devices",
+ "▁pict ures",
+ "▁picture s",
+ "▁t ut",
+ "▁tu t",
+ "\" `",
+ "▁im possible",
+ "▁r ail",
+ "▁ra il",
+ "▁fe els",
+ "▁feel s",
+ "▁fee ls",
+ "ic as",
+ "ica s",
+ "i cas",
+ "il ling",
+ "ill ing",
+ "▁acc ident",
+ "▁' @",
+ "____ ____",
+ "▁n otes",
+ "▁not es",
+ "▁no tes",
+ "▁note s",
+ "▁ notes",
+ "om an",
+ "oma n",
+ "o man",
+ "Par ser",
+ "Parse r",
+ "Pars er",
+ "▁dis covered",
+ "▁discover ed",
+ "▁R oman",
+ "▁Rom an",
+ "▁Ro man",
+ "▁Roma n",
+ "▁bud get",
+ "▁gu ide",
+ "▁guid e",
+ "ki ng",
+ "kin g",
+ "k ing",
+ "▁in cred",
+ "▁inc red",
+ "▁incre d",
+ "ol ar",
+ "ola r",
+ "o lar",
+ "en den",
+ "end en",
+ "ende n",
+ "Des c",
+ "De sc",
+ "D esc",
+ "▁w ave",
+ "▁wa ve",
+ "▁ wave",
+ "б ли",
+ "ig t",
+ "i gt",
+ "▁re strict",
+ "▁rest rict",
+ "▁restr ict",
+ "▁R et",
+ "▁Re t",
+ "▁ Ret",
+ "▁m ac",
+ "▁ma c",
+ "▁ mac",
+ "у р",
+ "B S",
+ "í s",
+ "▁gener ation",
+ "de m",
+ "d em",
+ "al o",
+ "a lo",
+ "б ра",
+ "▁order ed",
+ "▁ord ered",
+ "▁ ordered",
+ "dr op",
+ "dro p",
+ "d rop",
+ "▁p p",
+ "▁ pp",
+ "▁Re view",
+ "▁Rev iew",
+ "▁ Review",
+ "▁liter ally",
+ "▁literal ly",
+ "▁S ir",
+ "▁Si r",
+ "▁ Sir",
+ "▁Y eah",
+ "▁Ye ah",
+ "▁ Yeah",
+ "▁d ensity",
+ "▁dens ity",
+ "▁ density",
+ "ri z",
+ "r iz",
+ "in de",
+ "ind e",
+ "i nde",
+ "▁g ain",
+ "▁ga in",
+ "▁ gain",
+ "▁p anel",
+ "▁pan el",
+ "▁pa nel",
+ "▁ panel",
+ "je t",
+ "j et",
+ "▁T imes",
+ "▁Time s",
+ "▁Tim es",
+ "▁Ti mes",
+ "▁ Times",
+ "▁n ella",
+ "▁ne lla",
+ "▁nel la",
+ "▁nell a",
+ "▁pre viously",
+ "▁previous ly",
+ "▁prev iously",
+ "point s",
+ "Se nd",
+ "S end",
+ "▁B rown",
+ "▁Br own",
+ "▁Bro wn",
+ "▁Brow n",
+ "ea ch",
+ "e ach",
+ "▁tr igger",
+ "▁ trigger",
+ "ome times",
+ "omet imes",
+ "ic os",
+ "ico s",
+ "i cos",
+ "G R",
+ "Pane l",
+ "Pan el",
+ "P anel",
+ "og en",
+ "oge n",
+ "o gen",
+ "▁c m",
+ "▁ cm",
+ "ru ctions",
+ "ruct ions",
+ "ruction s",
+ "▁k iss",
+ "▁ki ss",
+ "▁s olo",
+ "▁so lo",
+ "▁sol o",
+ "▁f amous",
+ "▁fam ous",
+ "ra n",
+ "r an",
+ "п ро",
+ "▁th ro",
+ "▁thr o",
+ "Gr aph",
+ "G raph",
+ "im it",
+ "imi t",
+ "i mit",
+ "▁V alue",
+ "▁Val ue",
+ "▁ Value",
+ "▁st arts",
+ "▁start s",
+ "▁star ts",
+ "ip eline",
+ "ipe line",
+ "h d",
+ "T C",
+ "▁dis cussion",
+ "▁discuss ion",
+ "▁tr uck",
+ "ak a",
+ "a ka",
+ "On ly",
+ "▁E qu",
+ "▁Eq u",
+ "▁ Equ",
+ "▁k ö",
+ "▁ kö",
+ "▁B es",
+ "▁Be s",
+ "▁crit ic",
+ "▁pro pos",
+ "▁prop os",
+ "▁b att",
+ "▁bat t",
+ "▁ba tt",
+ "▁S ection",
+ "▁Se ction",
+ "▁ Section",
+ "Sh ow",
+ "S how",
+ "g p",
+ "ST ATE",
+ "STAT E",
+ "PO ST",
+ "POS T",
+ "P OST",
+ "▁N ord",
+ "▁No rd",
+ "▁Nor d",
+ "▁in nov",
+ "▁inn ov",
+ "▁c rim",
+ "▁cr im",
+ "▁cri m",
+ "▁ crim",
+ "ax is",
+ "a xis",
+ "▁T urn",
+ "▁Tur n",
+ "▁Tu rn",
+ "▁ Turn",
+ "con n",
+ "co nn",
+ "Run time",
+ "▁rem aining",
+ "▁remain ing",
+ "os ton",
+ "ost on",
+ "osto n",
+ "o ston",
+ "▁ Э",
+ "▁window s",
+ "▁wind ows",
+ "▁ windows",
+ "▁R oyal",
+ "▁Ro yal",
+ "▁Roy al",
+ "▁v ide",
+ "▁vi de",
+ "▁vid e",
+ "P P",
+ "ch ron",
+ "chr on",
+ "▁s an",
+ "▁sa n",
+ "▁ san",
+ "▁r ise",
+ "▁ri se",
+ "▁ris e",
+ "▁ rise",
+ "▁d elle",
+ "▁de lle",
+ "▁del le",
+ "▁dell e",
+ "▁D ur",
+ "▁Du r",
+ "▁rap id",
+ "▁ra pid",
+ "ce rt",
+ "cer t",
+ "c ert",
+ "L A",
+ "ed ge",
+ "▁\\ ]",
+ "▁ \\]",
+ "▁en tered",
+ "▁ent ered",
+ "▁enter ed",
+ "▁l aws",
+ "▁la ws",
+ "▁law s",
+ "▁ph oto",
+ "▁phot o",
+ "▁ photo",
+ "▁ap plications",
+ "▁applic ations",
+ "▁application s",
+ "▁appl ications",
+ "▁Ber lin",
+ "▁ar rest",
+ "▁arr est",
+ "▁f ederal",
+ "▁fed eral",
+ "▁feder al",
+ "▁R ussia",
+ "▁Russ ia",
+ "▁us ual",
+ "▁r aw",
+ "▁ra w",
+ "▁ raw",
+ "▁pi ù",
+ "êt re",
+ "ê tre",
+ "JS ON",
+ "J SON",
+ "SI ON",
+ "S ION",
+ "xt ure",
+ "ist ent",
+ "iste nt",
+ "isten t",
+ "▁P ower",
+ "▁Po wer",
+ "▁Pow er",
+ "▁ Power",
+ "Bi t",
+ "B it",
+ "▁cap acity",
+ "▁capac ity",
+ "▁ capacity",
+ "▁c ards",
+ "▁car ds",
+ "▁card s",
+ "▁ cards",
+ "UI D",
+ "U ID",
+ "im ents",
+ "iment s",
+ "imen ts",
+ "i ments",
+ "▁d ar",
+ "▁da r",
+ "▁ dar",
+ "▁Ch icago",
+ "▁comfort able",
+ "ti p",
+ "t ip",
+ "ba s",
+ "b as",
+ "▁m u",
+ "▁ mu",
+ "▁en emy",
+ "▁enem y",
+ "ya n",
+ "y an",
+ "▁ф и",
+ "▁ фи",
+ "▁up dated",
+ "▁update d",
+ "▁ updated",
+ "an go",
+ "ang o",
+ "E v",
+ "E ffect",
+ "os ing",
+ "osi ng",
+ "o sing",
+ "ren ce",
+ "r ence",
+ "▁Con gress",
+ "▁Cong ress",
+ "▁d efe",
+ "▁de fe",
+ "▁def e",
+ "▁i p",
+ "▁ ip",
+ "▁t out",
+ "▁to ut",
+ "▁tou t",
+ "▁f reedom",
+ "▁free dom",
+ "▁freed om",
+ "▁a o",
+ "▁ ao",
+ "▁There fore",
+ "▁Ther efore",
+ "Ed it",
+ "E dit",
+ "▁Vir gin",
+ "RE E",
+ "R EE",
+ "ar go",
+ "arg o",
+ "▁D am",
+ "▁Da m",
+ "▁ Dam",
+ "▁tra ffic",
+ "▁traff ic",
+ "ño s",
+ "ñ os",
+ "▁a lle",
+ "▁al le",
+ "▁all e",
+ "▁ alle",
+ "▁dep th",
+ "▁ depth",
+ "No w",
+ "N ow",
+ "▁s ides",
+ "▁side s",
+ "▁si des",
+ "▁sid es",
+ "▁го ди",
+ "▁год и",
+ "Des criptor",
+ "▁art ikel",
+ "▁n arrow",
+ "▁narr ow",
+ "▁nar row",
+ "__ _",
+ "_ __",
+ "k w",
+ "ut o",
+ "u to",
+ "▁Face book",
+ "▁Fac ebook",
+ "te gr",
+ "t egr",
+ "bo olean",
+ "ni k",
+ "n ik",
+ "b d",
+ "Tr ack",
+ "Tra ck",
+ "▁g ran",
+ "▁gr an",
+ "▁gra n",
+ "res hold",
+ "resh old",
+ "ве т",
+ "в ет",
+ "wr ap",
+ "w rap",
+ "▁n oise",
+ "▁no ise",
+ "ig u",
+ "i gu",
+ "▁B on",
+ "▁Bo n",
+ "▁ Bon",
+ "▁w y",
+ "▁ wy",
+ "lin ux",
+ "ck s",
+ "c ks",
+ "▁f ans",
+ "▁fa ns",
+ "▁fan s",
+ "▁m ach",
+ "▁ma ch",
+ "▁mac h",
+ "▁p rices",
+ "▁pr ices",
+ "▁pri ces",
+ "▁price s",
+ "é v",
+ "ou ts",
+ "out s",
+ "o uts",
+ "stand ing",
+ "stan ding",
+ "▁c ateg",
+ "▁cat eg",
+ "; \\",
+ "▁de cre",
+ "▁dec re",
+ "▁S aturday",
+ "▁m enu",
+ "▁me nu",
+ "▁men u",
+ "▁ menu",
+ "▁N ov",
+ "▁No v",
+ "▁Y et",
+ "▁Ye t",
+ "▁та к",
+ "lic he",
+ "li che",
+ "lich e",
+ "l iche",
+ "▁Ac adem",
+ "▁commun ication",
+ "us ing",
+ "u sing",
+ "▁Soc iety",
+ "▁Soci ety",
+ "▁n uc",
+ "▁nu c",
+ "pect ive",
+ "or ial",
+ "oria l",
+ "ori al",
+ "o rial",
+ "▁af raid",
+ "▁an imal",
+ "▁anim al",
+ "▁turn ing",
+ "▁tur ning",
+ "ds t",
+ "d st",
+ "math frak",
+ "le rs",
+ "ler s",
+ "l ers",
+ "▁l ots",
+ "▁lo ts",
+ "▁lot s",
+ "▁ á",
+ "▁T ra",
+ "▁Tr a",
+ "▁ Tra",
+ "n p",
+ "▁r ose",
+ "▁ro se",
+ "▁ rose",
+ "▁G L",
+ "▁ GL",
+ "▁hel ping",
+ "▁help ing",
+ "▁w inter",
+ "▁win ter",
+ "▁ко м",
+ "▁ ком",
+ "Mo ck",
+ "M ock",
+ "▁invest ment",
+ "Us e",
+ "U se",
+ "▁Can ad",
+ "н д",
+ "Co py",
+ "Cop y",
+ "C opy",
+ "▁f ly",
+ "▁fl y",
+ "▁ fly",
+ "SE R",
+ "S ER",
+ "▁F ar",
+ "▁Fa r",
+ "▁R os",
+ "▁Ro s",
+ "am il",
+ "ami l",
+ "a mil",
+ "▁fight ing",
+ "▁rel igious",
+ "▁relig ious",
+ "su per",
+ "sup er",
+ "s uper",
+ "sc reen",
+ "scr een",
+ "s creen",
+ "▁f urn",
+ "▁fur n",
+ "▁fu rn",
+ "▁surpr ised",
+ "▁surprise d",
+ "▁re plied",
+ "▁repl ied",
+ "Act ivity",
+ "Activ ity",
+ "▁D own",
+ "▁Do wn",
+ "▁Dow n",
+ "▁ Down",
+ "▁in sert",
+ "▁ins ert",
+ "▁ insert",
+ "▁O lymp",
+ "▁point ed",
+ "▁po inted",
+ "▁C ard",
+ "▁Car d",
+ "▁Ca rd",
+ "▁ Card",
+ "dr iver",
+ "drive r",
+ "d river",
+ "▁D a",
+ "▁ Da",
+ "! --",
+ "ro ud",
+ "rou d",
+ "r oud",
+ "un do",
+ "und o",
+ "▁m essages",
+ "▁message s",
+ "▁mess ages",
+ "▁ messages",
+ "▁P oint",
+ "▁Po int",
+ "▁ Point",
+ "V M",
+ "▁p lane",
+ "▁pl ane",
+ "▁plan e",
+ "▁ plane",
+ "x c",
+ "▁telev ision",
+ "▁tele vision",
+ "▁televis ion",
+ "ё н",
+ "▁thous ands",
+ "▁thousand s",
+ "▁c ris",
+ "▁cr is",
+ "▁cri s",
+ "▁de lay",
+ "▁del ay",
+ "▁ delay",
+ "▁N ext",
+ "▁Ne xt",
+ "▁ Next",
+ "▁no mbre",
+ "▁nom bre",
+ "▁t u",
+ "▁ tu",
+ "▁sk ip",
+ "▁ski p",
+ "▁ skip",
+ "ro ad",
+ "r oad",
+ "istr ation",
+ "▁t ur",
+ "▁tu r",
+ "▁De velop",
+ "▁Devel op",
+ "▁П а",
+ "▁д ру",
+ "▁др у",
+ "▁wonder ful",
+ "> &",
+ "▁L iber",
+ "▁Li ber",
+ "▁Lib er",
+ "▁s cope",
+ "▁sc ope",
+ "▁ scope",
+ "▁man age",
+ "▁ma nage",
+ "▁d ass",
+ "▁da ss",
+ "▁das s",
+ "▁re call",
+ "▁rec all",
+ "P M",
+ "▁re levant",
+ "▁relev ant",
+ "▁E arth",
+ "▁ка к",
+ "▁a pr",
+ "▁ap r",
+ "▁A SS",
+ "▁AS S",
+ "▁ ASS",
+ "ié n",
+ "i én",
+ "▁S H",
+ "▁ SH",
+ "oo m",
+ "o om",
+ "it et",
+ "ite t",
+ "no ne",
+ "non e",
+ "n one",
+ "as i",
+ "a si",
+ "▁mot or",
+ "▁mo tor",
+ "▁S how",
+ "▁Sh ow",
+ "▁ Show",
+ "n b",
+ "▁fact ors",
+ "▁fa ctors",
+ "▁factor s",
+ "▁f orest",
+ "▁for est",
+ "▁fore st",
+ "▁fo rest",
+ "▁в ре",
+ "th m",
+ "t hm",
+ "▁m unicip",
+ "▁turn s",
+ "▁tur ns",
+ "▁Div ision",
+ "▁Di vision",
+ "E C",
+ "▁dis appe",
+ "struct or",
+ "stru ctor",
+ "▁some where",
+ "▁Afr ican",
+ "▁Africa n",
+ "▁Inst itute",
+ "▁Institut e",
+ "Gr id",
+ "G rid",
+ "▁te acher",
+ "▁teach er",
+ "▁tea cher",
+ "ur ies",
+ "uri es",
+ "u ries",
+ "▁respect ively",
+ "▁respective ly",
+ "▁S D",
+ "▁ SD",
+ "▁a live",
+ "▁al ive",
+ "▁ali ve",
+ "▁p ou",
+ "▁po u",
+ "▁W ater",
+ "▁Wat er",
+ "▁Wa ter",
+ "▁ Water",
+ "ф е",
+ "▁ch anging",
+ "▁chang ing",
+ "▁ changing",
+ "▁after noon",
+ "▁or ders",
+ "▁order s",
+ "▁ord ers",
+ "▁ orders",
+ "Re t",
+ "R et",
+ "Point er",
+ "Po inter",
+ "▁s av",
+ "▁sa v",
+ "er g",
+ "e rg",
+ "ok ed",
+ "oke d",
+ "o ked",
+ "ess ions",
+ "ession s",
+ "▁F ire",
+ "▁Fi re",
+ "▁ Fire",
+ "ar et",
+ "are t",
+ "a ret",
+ "im m",
+ "i mm",
+ "▁des ire",
+ "▁ що",
+ "▁De sign",
+ "▁Des ign",
+ "▁ Design",
+ "ut ure",
+ "▁Off ice",
+ "▁c md",
+ "▁cm d",
+ "▁ cmd",
+ "▁e ating",
+ "▁eat ing",
+ "Net work",
+ "▁r ough",
+ "▁ro ugh",
+ "▁rou gh",
+ "▁ rough",
+ "oper ator",
+ "IG N",
+ "I GN",
+ "▁s ports",
+ "▁sp orts",
+ "▁sport s",
+ "▁w eren",
+ "▁we ren",
+ "▁were n",
+ "▁wer en",
+ "▁n oted",
+ "▁not ed",
+ "▁no ted",
+ "▁note d",
+ "▁tw ice",
+ "II I",
+ "I II",
+ "▁a nx",
+ "▁an x",
+ "▁e lim",
+ "▁el im",
+ "▁а в",
+ "▁i o",
+ "▁ io",
+ "▁spe ech",
+ "▁con du",
+ "▁cond u",
+ "el les",
+ "ell es",
+ "elle s",
+ "id ade",
+ "ida de",
+ "idad e",
+ "▁adv ance",
+ "R I",
+ "oc a",
+ "o ca",
+ "/ \\",
+ "ap shot",
+ "aps hot",
+ "▁t ail",
+ "▁ta il",
+ "▁ tail",
+ "mod els",
+ "model s",
+ "mode ls",
+ "og y",
+ "o gy",
+ "▁J eff",
+ "▁Je ff",
+ "ir ation",
+ "irat ion",
+ "▁K ore",
+ "▁Ko re",
+ "▁Kor e",
+ "▁le ads",
+ "▁lead s",
+ "ba t",
+ "b at",
+ "Ad apter",
+ "c ategory",
+ "ang ular",
+ "angu lar",
+ "▁s aved",
+ "▁sa ved",
+ "▁save d",
+ "▁sav ed",
+ "▁ saved",
+ "▁un iform",
+ "▁ uniform",
+ "▁n é",
+ "▁ né",
+ "▁business es",
+ "His t",
+ "Hi st",
+ "H ist",
+ "▁а р",
+ "▁ ар",
+ "do main",
+ "dom ain",
+ "▁S i",
+ "▁ Si",
+ "ra ise",
+ "rais e",
+ "rai se",
+ "r aise",
+ "▁w arn",
+ "▁war n",
+ "▁wa rn",
+ "▁ warn",
+ "het ic",
+ "h etic",
+ "▁G ro",
+ "▁Gr o",
+ ")) .",
+ ") ).",
+ "} >",
+ "з е",
+ "▁Amaz on",
+ "▁Or gan",
+ "▁ Organ",
+ "▁L ake",
+ "▁La ke",
+ "▁ag reement",
+ "▁agree ment",
+ "▁agre ement",
+ "x a",
+ "▁p erman",
+ "▁per man",
+ "▁perm an",
+ "▁cont aining",
+ "▁contain ing",
+ "▁st range",
+ "▁str ange",
+ "▁strang e",
+ "ст і",
+ "с ті",
+ "▁st upid",
+ "▁spe aking",
+ "▁speak ing",
+ "▁Intern et",
+ "▁Inter net",
+ "pre fix",
+ "pref ix",
+ "p refix",
+ "es c",
+ "e sc",
+ "As sert",
+ "Ass ert",
+ "pro te",
+ "pr ote",
+ "prot e",
+ "p rote",
+ "▁m anner",
+ "▁man ner",
+ "▁S z",
+ "un te",
+ "unt e",
+ "u nte",
+ "io t",
+ "i ot",
+ "Pro file",
+ "ov en",
+ "ove n",
+ "o ven",
+ "▁for med",
+ "▁form ed",
+ "▁forme d",
+ "▁ formed",
+ "▁l it",
+ "▁li t",
+ "▁ lit",
+ "▁econom y",
+ "▁ec onomy",
+ "▁c z",
+ "▁ cz",
+ "wi d",
+ "w id",
+ "RE Q",
+ "R EQ",
+ "▁ch osen",
+ "▁cho sen",
+ "▁chose n",
+ "▁P rodu",
+ "▁Pro du",
+ "▁ Produ",
+ "os ter",
+ "ost er",
+ "o ster",
+ "st ances",
+ "stance s",
+ "stan ces",
+ "aw a",
+ "a wa",
+ "▁R en",
+ "▁Re n",
+ "▁conf irm",
+ "▁ confirm",
+ "▁Б о",
+ "▁b illion",
+ "▁bill ion",
+ "▁d éc",
+ "▁dé c",
+ "ý ch",
+ "▁ill ustr",
+ "TI ES",
+ "T IES",
+ "▁P ub",
+ "▁Pu b",
+ "▁ Pub",
+ "▁b an",
+ "▁ba n",
+ "▁ ban",
+ "ad ed",
+ "ade d",
+ "a ded",
+ "ah n",
+ "a hn",
+ "▁C ath",
+ "▁Cat h",
+ "▁Ca th",
+ "no number",
+ "non umber",
+ "▁wor st",
+ "▁М е",
+ "▁sugg ested",
+ "▁suggest ed",
+ "st ats",
+ "stat s",
+ "sta ts",
+ "▁c ant",
+ "▁can t",
+ "▁ca nt",
+ "▁al ign",
+ "▁ali gn",
+ "▁ align",
+ "kap pa",
+ "k appa",
+ "▁h en",
+ "▁he n",
+ "▁ hen",
+ "▁in iti",
+ "▁init i",
+ "'] )",
+ "' ])",
+ "B I",
+ "▁g arden",
+ "▁gar den",
+ "▁gard en",
+ "▁sec ure",
+ "▁secur e",
+ "▁ secure",
+ "▁\\ [",
+ "▁ \\[",
+ "hand ler",
+ "handle r",
+ "el li",
+ "ell i",
+ "e lli",
+ "ld ots",
+ "l dots",
+ "se cut",
+ "sec ut",
+ "s ecut",
+ "▁ext ended",
+ "▁extend ed",
+ "} -",
+ "an ie",
+ "ani e",
+ "a nie",
+ "▁F ind",
+ "▁Fin d",
+ "▁Fi nd",
+ "▁ Find",
+ "▁M useum",
+ "▁Muse um",
+ "▁C onne",
+ "▁Con ne",
+ "▁ Conne",
+ "y y",
+ "▁pass ion",
+ "ak ers",
+ "ake rs",
+ "aker s",
+ "a kers",
+ "ah r",
+ "a hr",
+ "olog ies",
+ "ologie s",
+ "▁equ ation",
+ "▁eq uation",
+ "▁ equation",
+ "▁occ asion",
+ "▁occas ion",
+ "Le t",
+ "L et",
+ "'] ['",
+ "'][ '",
+ "' ]['",
+ "Pr int",
+ "an es",
+ "ane s",
+ "a nes",
+ "ie nte",
+ "ient e",
+ "ien te",
+ "i ente",
+ "▁T oday",
+ "▁To day",
+ "▁Tod ay",
+ "LE CT",
+ "L ECT",
+ "▁A f",
+ "▁ Af",
+ ", ,",
+ "▁Т а",
+ "▁` ``",
+ "▁`` `",
+ "ev en",
+ "eve n",
+ "e ven",
+ "si n",
+ "s in",
+ "ur er",
+ "ure r",
+ "u rer",
+ "▁ °",
+ "ot imes",
+ "oti mes",
+ "o times",
+ "▁I O",
+ "▁ IO",
+ "▁po et",
+ "() ));",
+ "()) );",
+ "())) ;",
+ "( )));",
+ "▁ −",
+ "▁ad opt",
+ "ph ere",
+ "pher e",
+ "p here",
+ "# [",
+ "▁c entre",
+ "▁cent re",
+ "ov es",
+ "ove s",
+ "o ves",
+ "▁a ns",
+ "▁an s",
+ "▁ ans",
+ "d p",
+ "▁K ir",
+ "▁Ki r",
+ "▁applic able",
+ "f p",
+ "▁vis ual",
+ "▁ok ay",
+ "or o",
+ "o ro",
+ "▁opportun ities",
+ "Re pository",
+ "Rep ository",
+ "▁l l",
+ "▁ ll",
+ "▁R od",
+ "▁Ro d",
+ "▁s hel",
+ "▁sh el",
+ "▁she l",
+ "▁la unch",
+ "▁con ven",
+ "▁conv en",
+ "▁conve n",
+ "▁S pe",
+ "▁Sp e",
+ "▁ Spe",
+ "Am er",
+ "A mer",
+ "▁c ette",
+ "▁cet te",
+ "Con d",
+ "Co nd",
+ "C ond",
+ "de p",
+ "d ep",
+ "O wn",
+ "▁h ook",
+ "▁ho ok",
+ "▁ hook",
+ "▁d ict",
+ "▁di ct",
+ "▁dic t",
+ "▁ dict",
+ "▁Th ose",
+ "▁f ellow",
+ "▁fell ow",
+ "▁fel low",
+ "▁phil osoph",
+ "▁philos oph",
+ "vi n",
+ "v in",
+ "fer ences",
+ "ference s",
+ "ha v",
+ "h av",
+ "▁ad ding",
+ "▁add ing",
+ "▁ adding",
+ "ivers e",
+ "iver se",
+ "i verse",
+ "ga me",
+ "g ame",
+ "▁Bl ue",
+ "▁ Blue",
+ "▁c lin",
+ "▁cl in",
+ "not e",
+ "no te",
+ "n ote",
+ "▁R am",
+ "▁Ra m",
+ "ме р",
+ "м ер",
+ "co very",
+ "cover y",
+ "cov ery",
+ "c overy",
+ "ñ a",
+ "▁б и",
+ "▁ би",
+ "▁f ashion",
+ "▁b roke",
+ "▁br oke",
+ "▁bro ke",
+ "▁' \\",
+ "▁ '\\",
+ "▁re ader",
+ "▁read er",
+ "▁ reader",
+ "но е",
+ "но сти",
+ "ност и",
+ "▁pay ment",
+ "▁ payment",
+ "▁L ic",
+ "▁Li c",
+ "▁l ips",
+ "▁li ps",
+ "▁lip s",
+ "▁ac adem",
+ "▁M ot",
+ "▁Mo t",
+ "el ls",
+ "ell s",
+ "C HECK",
+ "▁р у",
+ "▁ ру",
+ "▁M S",
+ "▁ MS",
+ "Ed itor",
+ "Edit or",
+ "▁z one",
+ "▁zo ne",
+ "▁ zone",
+ "it ure",
+ "itu re",
+ "▁I T",
+ "▁ IT",
+ "run time",
+ "▁pro ceed",
+ "▁proc eed",
+ "ло в",
+ "л ов",
+ "▁M aria",
+ "▁Mar ia",
+ "▁Ma ria",
+ "ol ver",
+ "olve r",
+ "olv er",
+ "▁Th anks",
+ "▁Thank s",
+ "▁ Thanks",
+ "▁should n",
+ "▁J oh",
+ "▁Jo h",
+ "▁Mod el",
+ "▁Mo del",
+ "▁Mode l",
+ "▁ Model",
+ "▁S ov",
+ "▁So v",
+ "! '",
+ "D i",
+ "▁c ancer",
+ "▁can cer",
+ "Id ent",
+ "▁ex change",
+ "il ler",
+ "ill er",
+ "ille r",
+ "in f",
+ "i nf",
+ "LE N",
+ "L EN",
+ "() {",
+ "( ){",
+ "ag a",
+ "a ga",
+ "\"] ,",
+ "\" ],",
+ "u h",
+ "▁K en",
+ "▁Ke n",
+ "▁ph otos",
+ "▁phot os",
+ "▁photo s",
+ "▁t iny",
+ "▁ti ny",
+ "▁tin y",
+ "▁ tiny",
+ "▁g ent",
+ "▁gen t",
+ "▁ge nt",
+ "▁ gent",
+ "ü l",
+ "▁T ake",
+ "▁Ta ke",
+ "▁Tak e",
+ "▁ Take",
+ "id el",
+ "ide l",
+ "i del",
+ "ou ting",
+ "out ing",
+ "In ternal",
+ "Inter nal",
+ "Intern al",
+ "▁c ells",
+ "▁cell s",
+ "▁cel ls",
+ "ни м",
+ "н им",
+ "ha rd",
+ "har d",
+ "h ard",
+ "▁T own",
+ "▁To wn",
+ "▁Tow n",
+ "ob e",
+ "o be",
+ "pl ex",
+ "ple x",
+ "p lex",
+ "те р",
+ "т ер",
+ "to ns",
+ "ton s",
+ "t ons",
+ "▁conc entr",
+ "▁concent r",
+ "mo ck",
+ "m ock",
+ "v c",
+ "á z",
+ "▁Ch ampionship",
+ "▁Champion ship",
+ "▁Champions hip",
+ "▁б е",
+ "▁ бе",
+ "? ?",
+ "ér i",
+ "é ri",
+ "al y",
+ "a ly",
+ "▁ Ц",
+ "ier te",
+ "iert e",
+ "▁tot ally",
+ "▁total ly",
+ "▁A uf",
+ "▁Au f",
+ "▁our selves",
+ "▁S elf",
+ "▁Sel f",
+ "▁ Self",
+ "Form s",
+ "For ms",
+ "ight er",
+ "igh ter",
+ "▁is land",
+ "fm t",
+ "f mt",
+ "▁r c",
+ "▁ rc",
+ "▁t ells",
+ "▁tell s",
+ "▁tel ls",
+ "B B",
+ "di t",
+ "d it",
+ "▁vari ables",
+ "▁variable s",
+ "▁ variables",
+ "▁int ended",
+ "▁intend ed",
+ "iz ont",
+ "izon t",
+ "izo nt",
+ "▁pl ays",
+ "▁play s",
+ "da m",
+ "d am",
+ "se q",
+ "s eq",
+ "▁S up",
+ "▁Su p",
+ "▁ Sup",
+ "▁c ultural",
+ "▁cult ural",
+ "▁sc ream",
+ "__ ,",
+ "_ _,",
+ "ci pl",
+ "cip l",
+ "Time out",
+ "▁ ж",
+ "or te",
+ "ort e",
+ "▁repl aced",
+ "▁replace d",
+ "E M",
+ "▁ab andon",
+ "▁Spec ial",
+ "▁Spe cial",
+ "▁ Special",
+ "el len",
+ "ell en",
+ "elle n",
+ "▁B ru",
+ "▁Br u",
+ "ir med",
+ "irm ed",
+ "T e",
+ "ol t",
+ "o lt",
+ "j u",
+ "Arg ument",
+ "▁ne ut",
+ "▁neu t",
+ "▁ neut",
+ "sc ape",
+ "▁R ay",
+ "▁Ra y",
+ "▁ Ray",
+ "▁Pol it",
+ "▁Po lit",
+ "▁crow d",
+ "▁cro wd",
+ "▁Window s",
+ "▁Wind ows",
+ "▁ Windows",
+ "ie go",
+ "ieg o",
+ "i ego",
+ "▁e scape",
+ "▁esc ape",
+ "▁ escape",
+ "▁Ap ache",
+ "sy nc",
+ "syn c",
+ "s ync",
+ "eb en",
+ "e ben",
+ "if ies",
+ "ifi es",
+ "et her",
+ "eth er",
+ "ethe r",
+ "e ther",
+ "Met a",
+ "Me ta",
+ "M eta",
+ "▁big gest",
+ "Ga me",
+ "G ame",
+ "▁trans action",
+ "▁ transaction",
+ "En v",
+ "E nv",
+ "▁М о",
+ "▁pl enty",
+ "▁m el",
+ "▁me l",
+ "▁ mel",
+ "п ре",
+ "▁mot iv",
+ "▁о р",
+ "▁ ор",
+ "or gan",
+ "org an",
+ "▁m ock",
+ "▁mo ck",
+ "▁ mock",
+ "▁$ _",
+ "▁ $_",
+ "ен е",
+ "е не",
+ "▁N umber",
+ "▁Num ber",
+ "▁Nu mber",
+ "▁ Number",
+ "ck now",
+ "c know",
+ "▁Up date",
+ "▁ Update",
+ "ze ro",
+ "zer o",
+ "z ero",
+ "▁sur prise",
+ "▁surpr ise",
+ "ce an",
+ "pd f",
+ "p df",
+ "Gl obal",
+ "▁att end",
+ "▁f ond",
+ "▁fo nd",
+ "▁fon d",
+ "▁under stood",
+ "Na v",
+ "N av",
+ "▁M ic",
+ "▁Mi c",
+ "▁ Mic",
+ "= $",
+ "ok ing",
+ "oki ng",
+ "o king",
+ "▁Stad ium",
+ "Cl ose",
+ "▁compet ition",
+ "▁sold iers",
+ "▁soldier s",
+ "▁O P",
+ "▁ OP",
+ "ag ne",
+ "agn e",
+ "▁An ton",
+ "▁Ant on",
+ "Ma in",
+ "M ain",
+ "á k",
+ "▁# [",
+ "▁ #[",
+ "▁Com mit",
+ "▁Comm it",
+ "▁ Commit",
+ "py x",
+ "▁e ast",
+ "▁eas t",
+ "▁ east",
+ "▁Or der",
+ "▁Ord er",
+ "▁ Order",
+ "F loat",
+ "▁accept ed",
+ "▁mon itor",
+ "▁ monitor",
+ "▁p ad",
+ "▁pa d",
+ "▁ pad",
+ "on ic",
+ "oni c",
+ "o nic",
+ "▁p ushed",
+ "▁push ed",
+ "▁re place",
+ "▁rep lace",
+ "▁repl ace",
+ "▁ replace",
+ "CR E",
+ "C RE",
+ "▁r ide",
+ "▁ri de",
+ "▁rid e",
+ "▁ ride",
+ "fo und",
+ "f ound",
+ "= %",
+ "во й",
+ "▁mat ches",
+ "▁match es",
+ "▁ matches",
+ "▁L ie",
+ "▁Li e",
+ "▁exper iences",
+ "▁experience s",
+ "▁experi ences",
+ "Po ol",
+ "P ool",
+ "up s",
+ "u ps",
+ "A V",
+ "▁ex istence",
+ "▁exist ence",
+ "▁t hin",
+ "▁th in",
+ "▁m agn",
+ "▁mag n",
+ "▁ma gn",
+ "CO MP",
+ "COM P",
+ "ho me",
+ "hom e",
+ "h ome",
+ "▁n i",
+ "▁ ni",
+ "▁wur den",
+ "▁wurde n",
+ "ла в",
+ "▁te eth",
+ "▁S tan",
+ "▁St an",
+ "▁Sta n",
+ "ap pro",
+ "app ro",
+ "an ny",
+ "ann y",
+ "if ts",
+ "ift s",
+ "▁un known",
+ "▁ unknown",
+ "▁h omes",
+ "▁home s",
+ "▁hom es",
+ "▁ho mes",
+ "▁ent ity",
+ "▁ entity",
+ "ci e",
+ "c ie",
+ "ле ние",
+ "ia r",
+ "i ar",
+ "▁compl iance",
+ "▁focus ed",
+ "uz z",
+ "u zz",
+ "=\\ \"",
+ "= \\\"",
+ "com ponents",
+ "component s",
+ "Att r",
+ "At tr",
+ "all ery",
+ "alle ry",
+ "aller y",
+ "▁ident ify",
+ "O k",
+ "pi e",
+ "p ie",
+ "▁St ill",
+ "▁off ering",
+ "▁offer ing",
+ "▁bu sy",
+ "▁bus y",
+ "ct l",
+ "c tl",
+ "it ors",
+ "itor s",
+ "ito rs",
+ "▁concern ed",
+ "▁concer ned",
+ "▁b rown",
+ "▁br own",
+ "▁bro wn",
+ "▁brow n",
+ "cl k",
+ "Se lected",
+ "Select ed",
+ "▁B lock",
+ "▁Bl ock",
+ "▁Blo ck",
+ "▁ Block",
+ "▁e gy",
+ "▁eg y",
+ "▁ egy",
+ "ic ing",
+ "ici ng",
+ "i cing",
+ "▁U RL",
+ "▁ URL",
+ "▁t opic",
+ "▁to pic",
+ "▁top ic",
+ "▁ topic",
+ "▁Pro duct",
+ "▁Produ ct",
+ "▁ Product",
+ "▁ч и",
+ "▁ чи",
+ "▁t rial",
+ "▁tr ial",
+ "▁tri al",
+ "▁week end",
+ "l u",
+ "▁I V",
+ "▁ IV",
+ "▁E gy",
+ "▁Eg y",
+ "x C",
+ "▁n ove",
+ "▁no ve",
+ "▁nov e",
+ "▁l ett",
+ "▁le tt",
+ "▁let t",
+ "▁ lett",
+ "en ne",
+ "enn e",
+ "() ).",
+ "()) .",
+ "( )).",
+ ".* *",
+ ". **",
+ "▁p romise",
+ "▁prom ise",
+ "el ection",
+ "ele ction",
+ "elect ion",
+ "e lection",
+ "Aut h",
+ "A uth",
+ "r v",
+ "ri l",
+ "r il",
+ "▁con duct",
+ "▁cond uct",
+ "▁condu ct",
+ "▁ conduct",
+ "▁main tain",
+ "▁maint ain",
+ "▁bo at",
+ "▁ boat",
+ "▁op posite",
+ "▁oppos ite",
+ "sp in",
+ "spi n",
+ "s pin",
+ "web pack",
+ "an ta",
+ "ant a",
+ "▁o rient",
+ "▁or ient",
+ "▁ orient",
+ "▁s uc",
+ "▁su c",
+ "▁ex ercise",
+ "▁exerc ise",
+ "▁eff icient",
+ "▁ efficient",
+ "▁trad ition",
+ "▁z w",
+ "▁ zw",
+ "▁S ud",
+ "▁Su d",
+ "go ing",
+ "▁P ier",
+ "▁Pi er",
+ "in v",
+ "i nv",
+ "ip es",
+ "ipe s",
+ "i pes",
+ "ensure math",
+ "▁con ver",
+ "▁conv er",
+ "▁conve r",
+ "cre en",
+ "cr een",
+ "c reen",
+ "▁t error",
+ "▁ter ror",
+ "▁terr or",
+ "▁D ou",
+ "▁Do u",
+ "▁in valid",
+ "▁ invalid",
+ "ce ived",
+ "ceive d",
+ "▁A rab",
+ "▁Ar ab",
+ "▁w ire",
+ "▁wir e",
+ "▁ wire",
+ "ap plication",
+ "sh ift",
+ "Gener ic",
+ "▁P lan",
+ "▁Pl an",
+ "▁ Plan",
+ "▁W all",
+ "▁Wal l",
+ "▁Wa ll",
+ "▁ Wall",
+ "▁direct ory",
+ "▁director y",
+ "▁ directory",
+ "▁e gg",
+ "▁eg g",
+ "▁we alth",
+ "▁ wealth",
+ "ran dom",
+ "rand om",
+ "r andom",
+ "att ribute",
+ "▁h ide",
+ "▁hi de",
+ "▁hid e",
+ "▁ hide",
+ "Se rial",
+ "Ser ial",
+ "S erial",
+ "ca m",
+ "c am",
+ "▁it al",
+ "▁i tal",
+ "▁ ital",
+ "▁L ine",
+ "▁Lin e",
+ "▁Li ne",
+ "▁ Line",
+ "▁C HECK",
+ "▁ CHECK",
+ "ploy ment",
+ "▁mass ive",
+ "▁ex tract",
+ "▁ext ract",
+ "▁extra ct",
+ "▁extr act",
+ "▁ extract",
+ "ch ain",
+ "cha in",
+ "Res t",
+ "Re st",
+ "R est",
+ "▁L as",
+ "▁La s",
+ "▁b ear",
+ "▁be ar",
+ "▁ bear",
+ "▁l inks",
+ "▁link s",
+ "▁lin ks",
+ "▁ links",
+ "▁new sp",
+ "▁news p",
+ "▁F C",
+ "▁ FC",
+ "Car d",
+ "C ard",
+ "ak s",
+ "a ks",
+ "▁v isible",
+ "▁vis ible",
+ "▁ visible",
+ "▁M arc",
+ "▁Mar c",
+ "▁Ma rc",
+ "▁B oston",
+ "▁Bo ston",
+ "▁Bos ton",
+ "▁res erved",
+ "▁reserv ed",
+ "▁reserve d",
+ "▁ro of",
+ "lic enses",
+ "license s",
+ "d c",
+ "▁In formation",
+ "▁ Information",
+ "▁w itness",
+ "S k",
+ "*) ,",
+ "* ),",
+ "Sc ope",
+ "S cope",
+ "'] ;",
+ "' ];",
+ "▁M ir",
+ "▁Mi r",
+ "▁ Mir",
+ "ud ing",
+ "udi ng",
+ "u ding",
+ "▁t rend",
+ "▁tr end",
+ "▁tre nd",
+ "▁tren d",
+ "re p",
+ "r ep",
+ "▁mus ical",
+ "▁music al",
+ "▁ne ither",
+ "▁nei ther",
+ "▁C reat",
+ "▁Cre at",
+ "▁ Creat",
+ "▁pos itions",
+ "▁position s",
+ "▁posit ions",
+ "L C",
+ "rid ge",
+ "r idge",
+ "▁offic ers",
+ "▁office rs",
+ "▁officer s",
+ "▁vi olence",
+ "▁viol ence",
+ "▁T em",
+ "▁Te m",
+ "▁S us",
+ "▁Su s",
+ "▁W ay",
+ "▁Wa y",
+ "Af ter",
+ "A fter",
+ "ac ket",
+ "ack et",
+ "▁S ou",
+ "▁So u",
+ "ac er",
+ "ace r",
+ "a cer",
+ "| |",
+ "▁re mark",
+ "▁r emark",
+ "▁rem ark",
+ "▁ remark",
+ "wa ter",
+ "w ater",
+ "n ě",
+ "▁С а",
+ "▁s ed",
+ "▁se d",
+ "▁ sed",
+ "E ach",
+ "▁phot ograph",
+ "▁photo graph",
+ "▁let ters",
+ "▁letter s",
+ "▁lett ers",
+ "▁in vent",
+ "▁inv ent",
+ "▁M as",
+ "▁Ma s",
+ "▁s ongs",
+ "▁son gs",
+ "▁song s",
+ "ó l",
+ "ki nd",
+ "kin d",
+ "k ind",
+ "▁N on",
+ "▁No n",
+ "▁ Non",
+ "▁d ust",
+ "▁du st",
+ "** :",
+ "* *:",
+ "nab la",
+ ".\" ,",
+ ". \",",
+ "Loc k",
+ "Lo ck",
+ "L ock",
+ "▁Д о",
+ "▁cl uster",
+ "▁ cluster",
+ "lo ss",
+ "los s",
+ "l oss",
+ "▁ASS ERT",
+ "▁ ASSERT",
+ "fa ll",
+ "f all",
+ "▁re ject",
+ "▁ reject",
+ "▁Sp ring",
+ "▁Spr ing",
+ "▁ Spring",
+ "▁wed ding",
+ "▁g rav",
+ "▁gr av",
+ "▁gra v",
+ "▁ grav",
+ "ress ion",
+ "r ession",
+ "li mit",
+ "lim it",
+ "l imit",
+ "RE S",
+ "R ES",
+ "] }",
+ "▁l isted",
+ "▁li sted",
+ "▁list ed",
+ "▁ listed",
+ "▁T ele",
+ "▁Te le",
+ "▁Tel e",
+ "▁ Tele",
+ "hl ine",
+ "h line",
+ "▁ch ief",
+ "▁chi ef",
+ "ME M",
+ "M EM",
+ "да р",
+ "д ар",
+ "▁exp ensive",
+ "tr ace",
+ "tra ce",
+ "▁R og",
+ "▁Ro g",
+ "▁C oll",
+ "▁Col l",
+ "▁Co ll",
+ "▁ Coll",
+ "▁Aut hor",
+ "▁Auth or",
+ "▁ Author",
+ "▁B oard",
+ "▁Bo ard",
+ "▁ Board",
+ "▁C apt",
+ "▁Cap t",
+ "▁Ca pt",
+ "▁ Capt",
+ "TE XT",
+ "T EXT",
+ "▁re con",
+ "▁rec on",
+ "es ta",
+ "est a",
+ "e sta",
+ "▁proper ly",
+ "▁& \\",
+ "▁ &\\",
+ "le ton",
+ "let on",
+ "l eton",
+ "ik er",
+ "ike r",
+ "i ker",
+ "G u",
+ "▁K om",
+ "▁Ko m",
+ "oc o",
+ "o co",
+ "▁any more",
+ "▁t aste",
+ "▁ta ste",
+ "▁tast e",
+ "▁S anta",
+ "▁San ta",
+ "▁Sant a",
+ "ge x",
+ "g ex",
+ "▁Se cret",
+ "▁Sec ret",
+ "▁ Secret",
+ "▁tal ent",
+ "▁tale nt",
+ "▁mom ents",
+ "▁moment s",
+ "▁mo ments",
+ "▁B a",
+ "▁ex tr",
+ "▁ext r",
+ "▁ extr",
+ "▁Com mission",
+ "▁Comm ission",
+ "▁mod ify",
+ "▁Fig ure",
+ "▁ Figure",
+ "▁d omin",
+ "▁do min",
+ "▁dom in",
+ "▁ domin",
+ "▁p lot",
+ "▁pl ot",
+ "▁ plot",
+ "en ger",
+ "eng er",
+ "enge r",
+ "ut ch",
+ "▁c ities",
+ "▁cit ies",
+ "▁ci ties",
+ "▁n ut",
+ "▁nu t",
+ "▁ nut",
+ "pro file",
+ "prof ile",
+ "▁S tat",
+ "▁St at",
+ "▁Sta t",
+ "▁ Stat",
+ "▁n odes",
+ "▁no des",
+ "▁node s",
+ "▁nod es",
+ "▁ nodes",
+ "▁n s",
+ "▁ ns",
+ "ess ages",
+ "essage s",
+ "essa ges",
+ "im pl",
+ "imp l",
+ "ic ker",
+ "ick er",
+ "i cker",
+ "▁ex amples",
+ "▁example s",
+ "▁exam ples",
+ "ab eth",
+ "abe th",
+ "abet h",
+ "▁st ated",
+ "▁stat ed",
+ "▁state d",
+ "▁sta ted",
+ "fi re",
+ "f ire",
+ "bu l",
+ "b ul",
+ "▁danger ous",
+ "▁P ay",
+ "▁Pa y",
+ "▁ Pay",
+ "▁G re",
+ "▁Gr e",
+ "▁ Gre",
+ "▁Mon day",
+ "▁Mond ay",
+ "es ome",
+ "eso me",
+ "e some",
+ "ig an",
+ "iga n",
+ "i gan",
+ "ru nd",
+ "run d",
+ "r und",
+ "pr ise",
+ "p rise",
+ "fa il",
+ "f ail",
+ "▁N ever",
+ "▁Ne ver",
+ "▁Nev er",
+ "▁ Never",
+ "A v",
+ "▁line ar",
+ "▁lin ear",
+ "▁ linear",
+ "▁u l",
+ "▁ ul",
+ "WA R",
+ "W AR",
+ "ре н",
+ "р ен",
+ "▁A T",
+ "▁ AT",
+ "▁d op",
+ "▁do p",
+ "▁n ou",
+ "▁no u",
+ "Des t",
+ "De st",
+ "D est",
+ "▁claim s",
+ "en da",
+ "end a",
+ "▁c razy",
+ "▁cr azy",
+ "ge l",
+ "g el",
+ "og gle",
+ "ogg le",
+ "▁rep resentation",
+ "▁represent ation",
+ "in en",
+ "ine n",
+ "i nen",
+ "▁altern ative",
+ "▁alter native",
+ "D M",
+ "AB ILITY",
+ "face s",
+ "fa ces",
+ "fac es",
+ "f aces",
+ "▁do ors",
+ "▁door s",
+ "▁ doors",
+ "at iv",
+ "ati v",
+ "Lo ok",
+ "L ook",
+ "▁J SON",
+ "▁JS ON",
+ "▁ JSON",
+ "▁appe arance",
+ "▁appear ance",
+ "б ря",
+ "S QL",
+ "▁sil ence",
+ "ud o",
+ "u do",
+ "▁Direct or",
+ "▁Dire ctor",
+ "▁Dir ector",
+ "State ment",
+ "Stat ement",
+ "se lected",
+ "select ed",
+ "hi gh",
+ "h igh",
+ "pr ime",
+ "prim e",
+ "▁ign ore",
+ "▁ignor e",
+ "▁ ignore",
+ "▁col ors",
+ "▁color s",
+ "▁ colors",
+ "us hing",
+ "ush ing",
+ "▁v irt",
+ "▁vi rt",
+ "▁vir t",
+ "▁ virt",
+ "man ager",
+ "▁rem ote",
+ "▁remot e",
+ "▁ remote",
+ "ł o",
+ "sm all",
+ "▁cr ime",
+ "▁crim e",
+ "▁cri me",
+ "r b",
+ "▁c reation",
+ "▁cre ation",
+ "▁creat ion",
+ "▁f light",
+ "▁fl ight",
+ "▁S ign",
+ "▁Si gn",
+ "▁Sig n",
+ "▁ Sign",
+ "IL E",
+ "I LE",
+ "▁D O",
+ "▁ DO",
+ "com ment",
+ "comm ent",
+ "▁C ost",
+ "▁Co st",
+ "▁Cos t",
+ "▁ Cost",
+ "._ _",
+ ". __",
+ "▁C op",
+ "▁Co p",
+ "▁ Cop",
+ "▁v om",
+ "▁vo m",
+ "▁Sc ience",
+ "▁Sci ence",
+ "ле ния",
+ "oo p",
+ "o op",
+ "inter face",
+ "▁WARRAN TIES",
+ "▁P age",
+ "▁Pa ge",
+ "▁ Page",
+ "** ****",
+ "**** **",
+ "*** ***",
+ "ско м",
+ "с ком",
+ "TR UE",
+ "▁re peated",
+ "▁repe ated",
+ "▁repeat ed",
+ "▁е го",
+ "ш о",
+ "▁r oz",
+ "▁ro z",
+ "▁ roz",
+ "P e",
+ "▁IS BN",
+ "ir ts",
+ "irt s",
+ "pos es",
+ "po ses",
+ "pose s",
+ "p oses",
+ "}) $",
+ "} )$",
+ "▁ І",
+ "child ren",
+ "ble s",
+ "bl es",
+ "b les",
+ "EC T",
+ "E CT",
+ "▁i z",
+ "▁ iz",
+ "▁b uilder",
+ "▁build er",
+ "▁ builder",
+ "▁M edia",
+ "▁Med ia",
+ "▁ Media",
+ "ia t",
+ "i at",
+ "▁contr ast",
+ "▁contra st",
+ "” ,",
+ "▁L ink",
+ "▁Lin k",
+ "▁ Link",
+ "▁Educ ation",
+ "▁j oint",
+ "▁join t",
+ "▁jo int",
+ "▁ joint",
+ "▁ex ternal",
+ "▁extern al",
+ "▁ external",
+ "▁ро з",
+ "▁b its",
+ "▁bit s",
+ "▁bi ts",
+ "▁ bits",
+ "FO RM",
+ "FOR M",
+ "F ORM",
+ "er man",
+ "erm an",
+ "w p",
+ "▁M ike",
+ "▁Mi ke",
+ "▁Mik e",
+ "▁M aster",
+ "▁Ma ster",
+ "▁Mas ter",
+ "▁ Master",
+ "▁sen ior",
+ "▁N av",
+ "▁Na v",
+ "▁ Nav",
+ "▁record ed",
+ "el ing",
+ "eli ng",
+ "elin g",
+ "e ling",
+ "es h",
+ "e sh",
+ "f x",
+ "ка н",
+ "к ан",
+ "▁t all",
+ "▁tal l",
+ "▁ta ll",
+ "▁John son",
+ "▁s ono",
+ "▁so no",
+ "▁son o",
+ "▁an che",
+ "▁anc he",
+ "▁anch e",
+ "▁ anche",
+ "ic ken",
+ "ick en",
+ "i cken",
+ "lo op",
+ "l oop",
+ "ici ency",
+ "empor ary",
+ "▁D oes",
+ "▁Do es",
+ "▁ Does",
+ "▁re lation",
+ "▁rel ation",
+ "▁ relation",
+ "м ы",
+ "wa s",
+ "w as",
+ "lo w",
+ "l ow",
+ "ich te",
+ "icht e",
+ "i chte",
+ "▁J ones",
+ "▁Jo nes",
+ "▁Jon es",
+ "▁bed room",
+ "DI S",
+ "D IS",
+ "▁mag net",
+ "▁magn et",
+ "▁Eng ine",
+ "▁ Engine",
+ "▁feel ings",
+ "▁feeling s",
+ "▁fee lings",
+ "G C",
+ "▁t orn",
+ "▁to rn",
+ "▁tor n",
+ "▁relationship s",
+ "▁relation ships",
+ "▁Р е",
+ "▁p roud",
+ "▁pro ud",
+ "▁pr oud",
+ "▁t we",
+ "▁tw e",
+ "ov al",
+ "ova l",
+ "o val",
+ "▁w aste",
+ "▁was te",
+ "▁wa ste",
+ "▁red uced",
+ "▁redu ced",
+ "▁reduce d",
+ "il ton",
+ "ilt on",
+ "B P",
+ "▁for got",
+ "▁forg ot",
+ "▁bod ies",
+ "▁H aw",
+ "▁Ha w",
+ "la g",
+ "l ag",
+ "▁w ww",
+ "▁ www",
+ "do or",
+ "d oor",
+ "▁s ufficient",
+ "▁suff icient",
+ "▁doll ars",
+ "▁dollar s",
+ "Le n",
+ "L en",
+ "▁talk ed",
+ "▁tal ked",
+ "▁b ond",
+ "▁bo nd",
+ "▁bon d",
+ "▁B or",
+ "▁Bo r",
+ "}} {",
+ "} }{",
+ "ro d",
+ "r od",
+ "Pass word",
+ "qu are",
+ "▁l ights",
+ "▁light s",
+ "▁ lights",
+ "er en",
+ "ere n",
+ "e ren",
+ "▁th irty",
+ "N C",
+ "▁T ODO",
+ "▁TO DO",
+ "▁res pond",
+ "▁respon d",
+ "▁resp ond",
+ "▁ respond",
+ "ки х",
+ "dir ect",
+ "di rect",
+ "dire ct",
+ "d irect",
+ "a ção",
+ "▁he av",
+ "Med ia",
+ "M edia",
+ "ex it",
+ "e xit",
+ "L icense",
+ "` .",
+ "▁m ixed",
+ "▁mix ed",
+ "▁d esk",
+ "▁de sk",
+ "▁des k",
+ "▁te aching",
+ "▁teach ing",
+ "▁tea ching",
+ "▁m aj",
+ "▁ma j",
+ "▁n erv",
+ "▁ne rv",
+ "▁ner v",
+ "in ations",
+ "ination s",
+ "type of",
+ "▁co ast",
+ "▁ж е",
+ "▁ же",
+ "▁be side",
+ "▁bes ide",
+ "um my",
+ "umm y",
+ "Do c",
+ "D oc",
+ "▁sche dule",
+ "▁schedul e",
+ "▁sched ule",
+ "▁ schedule",
+ "▁re cover",
+ "▁rec over",
+ "▁Fur ther",
+ "▁ste el",
+ "bo ot",
+ "b oot",
+ "▁Per haps",
+ "▁с ъ",
+ "▁O s",
+ "▁ Os",
+ "ri ck",
+ "ric k",
+ "r ick",
+ "▁В и",
+ "Supp ort",
+ "Sup port",
+ "S upport",
+ "▁( _",
+ "▁ (_",
+ "ni l",
+ "n il",
+ "pi s",
+ "p is",
+ "x pected",
+ "▁process ing",
+ "▁proces sing",
+ "▁ processing",
+ "Bu ild",
+ "B uild",
+ "ar ian",
+ "ari an",
+ "aria n",
+ "a rian",
+ "▁i con",
+ "▁ic on",
+ "▁ icon",
+ "▁C A",
+ "▁ CA",
+ "wi ck",
+ "w ick",
+ "= (",
+ "▁al gorithm",
+ "▁ algorithm",
+ "▁You ng",
+ "▁Man agement",
+ "▁ Management",
+ "▁anc ient",
+ "▁anci ent",
+ "но сть",
+ "ност ь",
+ "ot i",
+ "o ti",
+ "▁comb ination",
+ "wor ld",
+ "w orld",
+ "n n",
+ "▁d ram",
+ "▁dr am",
+ "en abled",
+ "ena bled",
+ "enable d",
+ "A c",
+ "C CESS",
+ "ar ation",
+ "▁bl ocks",
+ "▁block s",
+ "▁blo cks",
+ "▁ blocks",
+ "▁Ang eles",
+ "▁Angel es",
+ "▁Q ual",
+ "▁Qu al",
+ "▁ Qual",
+ "▁suc ceed",
+ "▁succ eed",
+ "net work",
+ "▁ob lig",
+ "spring framework",
+ "▁T re",
+ "▁Tr e",
+ "ok es",
+ "oke s",
+ "o kes",
+ "mu n",
+ "m un",
+ "▁Net work",
+ "▁ Network",
+ "De l",
+ "D el",
+ "▁e state",
+ "▁est ate",
+ "▁esta te",
+ "▁l iqu",
+ "▁li qu",
+ "▁p ob",
+ "▁po b",
+ "▁d ad",
+ "▁da d",
+ "▁dist inct",
+ "▁T it",
+ "▁Ti t",
+ "▁L ear",
+ "▁Le ar",
+ "fer red",
+ "and roid",
+ "andro id",
+ "▁sub sequ",
+ "▁subs equ",
+ "▁Flor ida",
+ "sub set",
+ "▁whis per",
+ "Vo l",
+ "V ol",
+ "ul ous",
+ "ulo us",
+ "▁c rew",
+ "▁cre w",
+ "▁cr ew",
+ "▁l ug",
+ "▁lu g",
+ "pi d",
+ "p id",
+ "oc ity",
+ "oci ty",
+ "o city",
+ "sk b",
+ "s kb",
+ "▁t ea",
+ "▁te a",
+ "у н",
+ "▁hon or",
+ "▁ho nor",
+ "▁I ns",
+ "▁In s",
+ "▁ Ins",
+ "▁g ew",
+ "▁ge w",
+ "▁ gew",
+ "Det ails",
+ "Detail s",
+ "ene ath",
+ "e neath",
+ "at ar",
+ "ata r",
+ "a tar",
+ "▁_ {",
+ "▁ _{",
+ "am en",
+ "ame n",
+ "a men",
+ "▁set up",
+ "▁ setup",
+ "Trans action",
+ "▁bl ank",
+ "▁ blank",
+ "Fail ed",
+ "F ailed",
+ "jo b",
+ "j ob",
+ "▁p ret",
+ "▁pre t",
+ "▁pr et",
+ "▁ pret",
+ "ß e",
+ "lo or",
+ "l oor",
+ "ř í",
+ "nc ia",
+ "n cia",
+ "▁any where",
+ "▁L ight",
+ "▁Li ght",
+ "▁ Light",
+ "▁A k",
+ "B D",
+ "▁exc ited",
+ "▁excit ed",
+ "ag ers",
+ "age rs",
+ "ager s",
+ "a gers",
+ "▁w arning",
+ "▁war ning",
+ "▁warn ing",
+ "▁ warning",
+ "▁process es",
+ "▁proces ses",
+ "h u",
+ "▁y outh",
+ "▁you th",
+ "▁yo uth",
+ "▁d ogs",
+ "▁do gs",
+ "▁dog s",
+ "▁o ct",
+ "▁oc t",
+ "▁ oct",
+ "▁n ine",
+ "▁ni ne",
+ "▁nin e",
+ "Write r",
+ "Wr iter",
+ "Writ er",
+ "W riter",
+ "gr id",
+ "g rid",
+ "▁import ance",
+ "est ic",
+ "▁care fully",
+ "▁careful ly",
+ "ma ster",
+ "mas ter",
+ "m aster",
+ "▁dec isions",
+ "▁decision s",
+ "▁decis ions",
+ "▁p in",
+ "▁pi n",
+ "▁ pin",
+ "▁cr ack",
+ "TE ST",
+ "TES T",
+ "T EST",
+ "▁L ocal",
+ "▁Loc al",
+ "▁Lo cal",
+ "▁ Local",
+ "▁R ight",
+ "▁ Right",
+ "▁v ast",
+ "▁va st",
+ "▁vas t",
+ "▁f aster",
+ "▁fa ster",
+ "▁fast er",
+ "▁inst itut",
+ "▁ann ual",
+ "LA N",
+ "L AN",
+ "▁e pisode",
+ "▁epis ode",
+ "▁X V",
+ "▁del ivery",
+ "▁deliver y",
+ "t l",
+ "F P",
+ "ci rc",
+ "cir c",
+ "▁typ ically",
+ "▁typical ly",
+ "ig o",
+ "i go",
+ "▁int el",
+ "▁inte l",
+ "▁ intel",
+ "na t",
+ "n at",
+ "x b",
+ "ст ро",
+ "с тро",
+ ") -",
+ "▁B al",
+ "▁Ba l",
+ "▁ Bal",
+ "▁J os",
+ "▁Jo s",
+ "▁g onna",
+ "▁R est",
+ "▁Re st",
+ "▁Res t",
+ "▁ Rest",
+ "jo r",
+ "j or",
+ "on ia",
+ "oni a",
+ "o nia",
+ "or ship",
+ "ors hip",
+ "ov ery",
+ "ove ry",
+ "over y",
+ "o very",
+ "LI NE",
+ "LIN E",
+ "L INE",
+ "] :",
+ "Que ue",
+ "▁com pare",
+ "▁comp are",
+ "▁compar e",
+ "▁ compare",
+ "▁ap artment",
+ "▁apart ment",
+ "▁r ul",
+ "▁ru l",
+ "D r",
+ "gen cy",
+ "g ency",
+ "▁ob viously",
+ "▁obvious ly",
+ "zi e",
+ "z ie",
+ "yc l",
+ "y cl",
+ "fort unately",
+ "fortun ately",
+ "fortunate ly",
+ "▁ste pped",
+ "▁step ped",
+ "▁S eg",
+ "▁Se g",
+ "▁ Seg",
+ "▁Wh ich",
+ "▁ Which",
+ "▁P C",
+ "▁ PC",
+ "▁a st",
+ "▁as t",
+ "▁ ast",
+ "end or",
+ "endo r",
+ "▁per mission",
+ "▁perm ission",
+ "▁ permission",
+ "CO L",
+ "C OL",
+ "▁T EST",
+ "▁TE ST",
+ "▁ TEST",
+ "P ay",
+ "ère s",
+ "è res",
+ "▁stud ied",
+ "▁accom pl",
+ "▁accomp l",
+ "ro le",
+ "rol e",
+ "r ole",
+ "Wh ere",
+ "Whe re",
+ "W here",
+ "proto buf",
+ "met adata",
+ "meta data",
+ "Jo b",
+ "J ob",
+ "▁F our",
+ "▁Fou r",
+ "▁Fo ur",
+ "pl ements",
+ "ple ments",
+ "plement s",
+ "dis able",
+ "▁l oud",
+ "▁lo ud",
+ "▁lou d",
+ "▁happ ening",
+ "▁happen ing",
+ "▁U sing",
+ "▁Us ing",
+ "▁ Using",
+ "ro g",
+ "r og",
+ "▁depend s",
+ "▁dep ends",
+ "í m",
+ "' \\",
+ "▁t aught",
+ "sh ared",
+ "sha red",
+ "share d",
+ "▁att ributes",
+ "▁attribute s",
+ "▁attribut es",
+ "▁ attributes",
+ "▁A ction",
+ "▁Act ion",
+ "▁ Action",
+ "▁d ess",
+ "▁de ss",
+ "▁des s",
+ "▁ dess",
+ "▁h ouses",
+ "▁house s",
+ "▁hous es",
+ "▁ho uses",
+ "▁re set",
+ "▁res et",
+ "▁ reset",
+ "▁b ien",
+ "▁bi en",
+ "▁ex plicit",
+ "▁expl icit",
+ "LO W",
+ "-> _",
+ "▁P M",
+ "▁ PM",
+ "C ategory",
+ "oi ce",
+ "o ice",
+ "in to",
+ "int o",
+ "▁m ail",
+ "▁ma il",
+ "▁mai l",
+ "▁ mail",
+ "▁author ity",
+ "▁un able",
+ "▁una ble",
+ "file name",
+ "fil ename",
+ "é k",
+ "ле й",
+ "л ей",
+ "▁s ector",
+ "▁se ctor",
+ "▁sec tor",
+ "▁sect or",
+ "ap point",
+ "app oint",
+ "▁h ang",
+ "▁ha ng",
+ "▁han g",
+ "▁ hang",
+ "▁c el",
+ "▁ce l",
+ "▁ cel",
+ "rel ated",
+ "it ate",
+ "ita te",
+ "itat e",
+ "▁' <",
+ "am ber",
+ "amb er",
+ "a mber",
+ "▁c heap",
+ "▁che ap",
+ "▁en abled",
+ "▁enable d",
+ "▁ enabled",
+ "▁di vision",
+ "▁div ision",
+ "▁divis ion",
+ "An y",
+ "A ny",
+ "▁h ier",
+ "▁hi er",
+ "▁H ead",
+ "▁He ad",
+ "▁ Head",
+ "nt ax",
+ "n tax",
+ "ud a",
+ "u da",
+ "▁lim itations",
+ "▁limit ations",
+ "▁limitation s",
+ "▁st udio",
+ "▁stud io",
+ "med ia",
+ "medi a",
+ "m edia",
+ "▁cir cle",
+ "▁circ le",
+ "▁ circle",
+ "но ва",
+ "нов а",
+ "▁l aug",
+ "▁la ug",
+ "ac ts",
+ "act s",
+ "▁В о",
+ "ó d",
+ "pl ed",
+ "ple d",
+ "p led",
+ "LO C",
+ "L OC",
+ "Ex pr",
+ "Exp r",
+ "> :",
+ "▁pr és",
+ "▁pré s",
+ "▁ prés",
+ "▁laugh ed",
+ "▁laug hed",
+ "▁Th ree",
+ "▁ Three",
+ "л ы",
+ "▁en ds",
+ "▁end s",
+ "▁ ends",
+ "▁fund ament",
+ "▁in her",
+ "▁ inher",
+ "▁l iv",
+ "▁li v",
+ "▁ liv",
+ "bi d",
+ "b id",
+ "▁respons ibility",
+ "▁check ed",
+ "▁ checked",
+ "▁P ac",
+ "▁Pa c",
+ "▁f ault",
+ "▁fa ult",
+ "▁y ellow",
+ "▁s alt",
+ "▁sa lt",
+ "▁sal t",
+ "▁Franc isco",
+ "▁Francis co",
+ "▁ ^",
+ "▁O N",
+ "▁ ON",
+ "▁beaut y",
+ "y g",
+ "▁A ff",
+ "▁Af f",
+ "▁ Aff",
+ "▁E q",
+ "▁ Eq",
+ "▁mag ic",
+ "▁hand ler",
+ "▁handle r",
+ "▁ handler",
+ "x E",
+ "▁numer ous",
+ "▁numero us",
+ "▁h ole",
+ "▁hol e",
+ "▁ho le",
+ "▁ hole",
+ "▁ro oms",
+ "▁room s",
+ "▁ rooms",
+ "cc ión",
+ "cció n",
+ "c ción",
+ "▁A rm",
+ "▁Ar m",
+ "▁ Arm",
+ "per son",
+ "pers on",
+ "p erson",
+ "▁build ings",
+ "▁building s",
+ "▁p late",
+ "▁pl ate",
+ "▁plat e",
+ "ble d",
+ "bl ed",
+ "b led",
+ "er rors",
+ "err ors",
+ "error s",
+ "▁A gain",
+ "▁Ag ain",
+ "▁Def ault",
+ "▁ Default",
+ "▁H ard",
+ "▁Har d",
+ "▁Ha rd",
+ "▁ Hard",
+ "t ó",
+ "hu s",
+ "h us",
+ "▁dim ension",
+ "ial e",
+ "ia le",
+ "i ale",
+ "▁M ult",
+ "▁Mu lt",
+ "▁Mul t",
+ "▁ Mult",
+ "▁Govern ment",
+ "Fun c",
+ "F unc",
+ "▁b low",
+ "▁bl ow",
+ "▁blo w",
+ "▁re ct",
+ "▁r ect",
+ "▁rec t",
+ "▁ rect",
+ "er ra",
+ "err a",
+ "conne ction",
+ "connect ion",
+ "conn ection",
+ "▁pass ing",
+ "▁pas sing",
+ "ße n",
+ "ß en",
+ "ph as",
+ "pha s",
+ "p has",
+ "ens ional",
+ "ension al",
+ "re cord",
+ "rec ord",
+ "co hol",
+ "▁H arry",
+ "▁Har ry",
+ "▁Harr y",
+ "izont al",
+ "izon tal",
+ "▁f inger",
+ "▁fin ger",
+ "▁fing er",
+ "▁young er",
+ "▁S C",
+ "▁ SC",
+ "oper ation",
+ "B Y",
+ "he im",
+ "▁B ad",
+ "▁Ba d",
+ "▁ Bad",
+ "▁st orm",
+ "▁stor m",
+ "▁sto rm",
+ "▁ storm",
+ "▁N at",
+ "▁Na t",
+ "▁bu ying",
+ "▁buy ing",
+ "▁S ometimes",
+ "▁Some times",
+ "▁С та",
+ "es sed",
+ "ess ed",
+ "esse d",
+ "▁da mn",
+ "▁dam n",
+ "▁m eg",
+ "▁me g",
+ "um es",
+ "ume s",
+ "u mes",
+ "ün d",
+ "ü nd",
+ "т ра",
+ "▁sil ver",
+ "w d",
+ "hid den",
+ "h idden",
+ "ar do",
+ "ard o",
+ "▁commun ities",
+ "▁d iet",
+ "▁di et",
+ "▁die t",
+ "ot ted",
+ "ott ed",
+ "otte d",
+ "▁b at",
+ "▁ba t",
+ "▁ bat",
+ "an cer",
+ "ance r",
+ "anc er",
+ "▁f mt",
+ "▁ fmt",
+ "▁P en",
+ "▁Pe n",
+ "▁ Pen",
+ "▁t il",
+ "▁ti l",
+ "▁ til",
+ "En um",
+ "E num",
+ "PA TH",
+ "P ATH",
+ "▁mat ters",
+ "▁matter s",
+ "▁matt ers",
+ "time out",
+ "-- ----------",
+ "---- --------",
+ "-------- ----",
+ "--- ---------",
+ "----- -------",
+ "---------- --",
+ "------ ------",
+ "--------- ---",
+ "------- -----",
+ "----------- -",
+ "- -----------",
+ "ka n",
+ "k an",
+ "▁Cor por",
+ "=\" ../../",
+ "=\"../ ../",
+ "▁A le",
+ "▁Al e",
+ "hent ication",
+ "hentic ation",
+ "▁com plic",
+ "▁comp lic",
+ "▁compl ic",
+ "▁Se curity",
+ "▁Sec urity",
+ "▁ Security",
+ "OF F",
+ "O FF",
+ "R ad",
+ "ap se",
+ "aps e",
+ "a pse",
+ "▁d ance",
+ "▁dan ce",
+ "▁perm issions",
+ "▁permission s",
+ "▁war rant",
+ "▁l ad",
+ "▁la d",
+ "▁ lad",
+ "▁is ol",
+ "▁i sol",
+ "d l",
+ "▁A u",
+ "ye s",
+ "y es",
+ "▁t v",
+ "▁ tv",
+ "▁pro vider",
+ "▁prov ider",
+ "▁provide r",
+ "▁ provider",
+ "▁ter rible",
+ "▁terr ible",
+ "▁dep artment",
+ "▁depart ment",
+ "er al",
+ "era l",
+ "e ral",
+ "▁implement ation",
+ "S R",
+ "▁h earing",
+ "▁he aring",
+ "▁hear ing",
+ "▁K n",
+ "F R",
+ "t v",
+ "▁d iss",
+ "▁dis s",
+ "▁di ss",
+ "F UN",
+ "▁dur ante",
+ "▁durant e",
+ "os is",
+ "osi s",
+ "o sis",
+ "▁task s",
+ "▁ tasks",
+ "▁B lo",
+ "▁Bl o",
+ "▁ Blo",
+ "во д",
+ "▁br anch",
+ "▁ branch",
+ "▁polit ics",
+ "▁E lle",
+ "▁El le",
+ "▁Ell e",
+ "▁lead ership",
+ "▁leader ship",
+ "▁leaders hip",
+ "ex pr",
+ "exp r",
+ "▁techn iques",
+ "▁technique s",
+ "pr ec",
+ "pre c",
+ "p rec",
+ "Sig ma",
+ "S igma",
+ "im ately",
+ "imate ly",
+ "imat ely",
+ "t k",
+ "ach ment",
+ "▁En ter",
+ "▁Ent er",
+ "▁ Enter",
+ "▁cre ative",
+ "▁creat ive",
+ "▁з на",
+ "▁ зна",
+ "ap py",
+ "app y",
+ "un ched",
+ "unch ed",
+ "unc hed",
+ "▁' ',",
+ "▁'' ,",
+ "on der",
+ "ond er",
+ "onde r",
+ "o nder",
+ "{ -",
+ "NU M",
+ "N UM",
+ "▁n arr",
+ "▁na rr",
+ "▁nar r",
+ "Mem ory",
+ "▁win ning",
+ "▁ winning",
+ "▁F ollow",
+ "▁Fol low",
+ "▁ Follow",
+ "*/ \r",
+ "vis ion",
+ "v ision",
+ "res ents",
+ "resent s",
+ "zi one",
+ "z ione",
+ "▁l atter",
+ "▁lat ter",
+ "▁requ ests",
+ "▁request s",
+ "▁ requests",
+ "▁m argin",
+ "▁mar gin",
+ "▁marg in",
+ "▁ margin",
+ "▁{ \"",
+ "▁ {\"",
+ "v ideo",
+ "c n",
+ "▁Im age",
+ "▁ Image",
+ "T im",
+ "CON FIG",
+ "CONF IG",
+ "▁all owing",
+ "▁allow ing",
+ "▁comb ined",
+ "▁combine d",
+ "PU T",
+ "P UT",
+ "▁instance of",
+ "ig in",
+ "igi n",
+ "i gin",
+ "▁p ero",
+ "▁per o",
+ "▁pe ro",
+ "▁' '",
+ "▁ ''",
+ "▁conf idence",
+ "▁equ ivalent",
+ "▁equival ent",
+ "pa d",
+ "p ad",
+ "ef fect",
+ "eff ect",
+ "e ffect",
+ "R X",
+ "▁l ang",
+ "▁la ng",
+ "▁lan g",
+ "▁ lang",
+ "str ong",
+ "▁b ridge",
+ "▁br idge",
+ "▁ bridge",
+ "ay a",
+ "a ya",
+ "▁t reated",
+ "▁tre ated",
+ "▁treat ed",
+ "▁f orth",
+ "▁for th",
+ "▁fort h",
+ "S W",
+ "▁account s",
+ "▁P O",
+ "▁ PO",
+ "▁list ening",
+ "▁listen ing",
+ "Ro ute",
+ "R oute",
+ "() ))",
+ "()) )",
+ "( )))",
+ "cp y",
+ "c py",
+ "▁re form",
+ "▁ref orm",
+ "▁g ate",
+ "▁ga te",
+ "▁ gate",
+ "▁W alk",
+ "▁Wal k",
+ "▁ Walk",
+ "▁some how",
+ "t f",
+ "▁l ayout",
+ "▁la yout",
+ "▁lay out",
+ "▁ layout",
+ "um in",
+ "umi n",
+ "u min",
+ "▁consider ing",
+ "▁consid ering",
+ "▁pre mi",
+ "▁pr emi",
+ "▁prem i",
+ "▁M om",
+ "▁Mo m",
+ "at han",
+ "ath an",
+ "a than",
+ "Ge n",
+ "G en",
+ "▁plan et",
+ "▁plane t",
+ "am ples",
+ "amp les",
+ "ample s",
+ "▁M O",
+ "▁ MO",
+ "sh op",
+ "s hop",
+ "▁prem ier",
+ "▁premi er",
+ "▁s impl",
+ "▁sim pl",
+ "▁s egu",
+ "▁se gu",
+ "▁seg u",
+ "L Y",
+ "Su m",
+ "S um",
+ "▁t ables",
+ "▁table s",
+ "▁tab les",
+ "▁ta bles",
+ "▁ tables",
+ "sk a",
+ "s ka",
+ "▁ ž",
+ "p d",
+ "▁s ous",
+ "▁so us",
+ "▁sou s",
+ "▁con ference",
+ "▁confer ence",
+ "▁D at",
+ "▁Da t",
+ "▁ Dat",
+ "Sc roll",
+ "▁stand ards",
+ "▁standard s",
+ "▁г ру",
+ "es se",
+ "ess e",
+ "▁citiz ens",
+ "▁citizen s",
+ "▁occur red",
+ "▁dem ocr",
+ "▁demo cr",
+ "▁e lev",
+ "▁el ev",
+ "▁ele v",
+ "▁S em",
+ "▁Se m",
+ "▁ Sem",
+ "ens us",
+ "he aders",
+ "head ers",
+ "header s",
+ "▁Ch ris",
+ "im ento",
+ "iment o",
+ "imen to",
+ "ko m",
+ "k om",
+ "Co r",
+ "C or",
+ "MI N",
+ "M IN",
+ "us her",
+ "ush er",
+ "Data base",
+ "Dat abase",
+ "▁f ormal",
+ "▁for mal",
+ "▁form al",
+ "▁forma l",
+ "ig ne",
+ "ign e",
+ "▁organ izations",
+ "▁organiz ations",
+ "▁organization s",
+ "▁I re",
+ "▁Ir e",
+ "X ml",
+ "и з",
+ "▁p ray",
+ "▁pr ay",
+ "▁pra y",
+ "▁b omb",
+ "▁bo mb",
+ "▁bom b",
+ "▁m and",
+ "▁man d",
+ "▁ma nd",
+ "▁ mand",
+ "er ts",
+ "ert s",
+ "▁c lock",
+ "▁cl ock",
+ "▁clo ck",
+ "▁ clock",
+ "▁b uck",
+ "▁bu ck",
+ "ва ли",
+ "вал и",
+ "в али",
+ "en sch",
+ "ens ch",
+ "▁v olt",
+ "▁vo lt",
+ "▁vol t",
+ "▁ volt",
+ "▁fil ms",
+ "▁film s",
+ "▁pl ants",
+ "▁plan ts",
+ "▁plant s",
+ "in ode",
+ "ino de",
+ "i node",
+ "Bo olean",
+ "▁restaur ant",
+ "ía n",
+ "í an",
+ "▁de but",
+ "▁deb ut",
+ "page s",
+ "pa ges",
+ "pag es",
+ "p ages",
+ "▁wor dt",
+ "▁word t",
+ "▁Б а",
+ "▁great est",
+ "(\" /",
+ "▁c opyright",
+ "▁copy right",
+ "▁ copyright",
+ "▁r it",
+ "▁ri t",
+ "▁ rit",
+ "size of",
+ "Tr ace",
+ "Tra ce",
+ "ue nt",
+ "uen t",
+ "u ent",
+ "ту р",
+ "т ур",
+ "▁k o",
+ "▁ ko",
+ ": \\",
+ "▁b igger",
+ "▁big ger",
+ "▁perfect ly",
+ "ten ance",
+ "MA SK",
+ "M ASK",
+ "r é",
+ "▁e tt",
+ "▁et t",
+ "▁ ett",
+ "▁n ose",
+ "▁no se",
+ "▁nos e",
+ "▁c raft",
+ "▁cr aft",
+ "▁ craft",
+ "it eral",
+ "ite ral",
+ "iter al",
+ "▁discuss ed",
+ "▁Jew ish",
+ "C ap",
+ "▁Un less",
+ "▁Jack son",
+ "Att ributes",
+ "Attribute s",
+ "Attrib utes",
+ "▁l unch",
+ "▁lun ch",
+ "ö l",
+ "at r",
+ "a tr",
+ "▁pay ing",
+ "▁pa ying",
+ "Par se",
+ "Pars e",
+ "P arse",
+ "() \r",
+ "( )\r",
+ "la d",
+ "l ad",
+ "▁r are",
+ "▁ra re",
+ "▁[ ];",
+ "▁[] ;",
+ "▁ [];",
+ "st one",
+ "ston e",
+ "sto ne",
+ "▁u nc",
+ "▁un c",
+ "▁ unc",
+ "▁def ense",
+ "▁defens e",
+ "} +",
+ "▁Gl obal",
+ "▁ Global",
+ "▁Sov iet",
+ "▁Austral ian",
+ "▁Australia n",
+ "▁g li",
+ "▁gl i",
+ "var iant",
+ "vari ant",
+ "▁R on",
+ "▁Ro n",
+ "▁lo an",
+ "St ep",
+ "Ste p",
+ "me mber",
+ "mem ber",
+ "m ember",
+ "Sc h",
+ "S ch",
+ "▁Commit tee",
+ "▁s pending",
+ "▁sp ending",
+ "▁spend ing",
+ "▁T ri",
+ "▁Tr i",
+ "▁ Tri",
+ "▁J ournal",
+ "▁Jour nal",
+ "▁ Journal",
+ "▁s ugar",
+ "▁su gar",
+ "▁sug ar",
+ "el ly",
+ "ell y",
+ "HT ML",
+ "▁ad vent",
+ "▁adv ent",
+ "win g",
+ "wi ng",
+ "w ing",
+ "▁Wh ether",
+ "▁Whe ther",
+ "or ation",
+ "▁N E",
+ "▁ NE",
+ "iv eness",
+ "ive ness",
+ "iven ess",
+ "▁h av",
+ "▁ha v",
+ "▁ hav",
+ "▁con scious",
+ "▁ conscious",
+ "ee n",
+ "e en",
+ "Sym bol",
+ "S ymbol",
+ "▁к у",
+ "▁ ку",
+ "Log ger",
+ "▁L ittle",
+ "▁Lit tle",
+ "wide t",
+ "wi det",
+ "wid et",
+ "oc ation",
+ "pi n",
+ "p in",
+ "▁sym met",
+ "▁A D",
+ "▁ AD",
+ "▁pos ts",
+ "▁po sts",
+ "▁post s",
+ "▁ posts",
+ "sh al",
+ "sha l",
+ "s hal",
+ "▁Con f",
+ "▁Co nf",
+ "▁ Conf",
+ "▁ch ose",
+ "▁cho se",
+ "ma l",
+ "m al",
+ "ul o",
+ "u lo",
+ "▁M ethod",
+ "▁ Method",
+ "▁miss ed",
+ "▁mis sed",
+ "Re move",
+ "Rem ove",
+ "Aut o",
+ "A uto",
+ "VAL UE",
+ "th let",
+ "▁For ce",
+ "▁ Force",
+ "p f",
+ "▁ Я",
+ "la te",
+ "lat e",
+ "l ate",
+ "▁p ul",
+ "▁pu l",
+ "▁ pul",
+ "Po p",
+ "P op",
+ "▁adv anced",
+ "▁advance d",
+ "air es",
+ "ai res",
+ "aire s",
+ "a ires",
+ "res sed",
+ "ress ed",
+ "resse d",
+ "r essed",
+ "AM E",
+ "A ME",
+ "be ll",
+ "bel l",
+ "b ell",
+ "ac hing",
+ "ach ing",
+ "achi ng",
+ "a ching",
+ "i ć",
+ "ec ho",
+ "ech o",
+ "e cho",
+ "H S",
+ "▁fun ny",
+ "ри и",
+ "▁e er",
+ "▁ve get",
+ "▁four th",
+ "c f",
+ "trans form",
+ "▁g rown",
+ "▁gr own",
+ "▁grow n",
+ "▁gro wn",
+ "▁Mc C",
+ "si te",
+ "s ite",
+ "▁b eneath",
+ "▁be neath",
+ "▁s hell",
+ "▁sh ell",
+ "▁she ll",
+ "▁shel l",
+ "▁ shell",
+ "x d",
+ "Pl ay",
+ "P lay",
+ "sh ort",
+ "Ro le",
+ "R ole",
+ "▁relig ion",
+ "in ator",
+ "ina tor",
+ "} ",
+ "▁El iz",
+ "▁Eli z",
+ "M icrosoft",
+ "▁v ez",
+ "▁ve z",
+ "▁ vez",
+ "▁ра бо",
+ "▁ рабо",
+ "re ich",
+ "rei ch",
+ "ve t",
+ "v et",
+ "en um",
+ "enu m",
+ "e num",
+ "▁w elcome",
+ "▁wel come",
+ "name nt",
+ "na ment",
+ "nam ent",
+ "n ament",
+ "▁j an",
+ "▁ja n",
+ "▁ jan",
+ "▁c ycle",
+ "▁cy cle",
+ "▁cycl e",
+ "▁ cycle",
+ "▁a cknow",
+ "▁ac know",
+ "▁w ound",
+ "▁wo und",
+ "id i",
+ "i di",
+ "▁poss ibility",
+ "an notation",
+ "annot ation",
+ "▁techn ical",
+ "▁f old",
+ "▁fol d",
+ "▁fo ld",
+ "▁ fold",
+ "e h",
+ "ist ence",
+ "isten ce",
+ "▁re ply",
+ "▁rep ly",
+ "▁repl y",
+ "▁ reply",
+ "et es",
+ "ete s",
+ "e tes",
+ "▁dec ades",
+ "▁decade s",
+ "wa n",
+ "w an",
+ "▁к ра",
+ "▁ кра",
+ "▁L ab",
+ "▁La b",
+ "▁u nf",
+ "▁un f",
+ "▁im per",
+ "▁imp er",
+ "▁ imper",
+ "▁b ug",
+ "▁bu g",
+ "▁ bug",
+ "▁Th ough",
+ "th rows",
+ "throw s",
+ "Vis ible",
+ "V isible",
+ "pr ev",
+ "pre v",
+ "p rev",
+ "▁T y",
+ "▁ Ty",
+ "▁de pending",
+ "▁depend ing",
+ "▁dep ending",
+ "▁pol icies",
+ "▁polic ies",
+ "an dy",
+ "and y",
+ "▁Ital ian",
+ "▁Italia n",
+ "um a",
+ "u ma",
+ "▁sign s",
+ "▁sig ns",
+ "▁Th rough",
+ "б ы",
+ "bo t",
+ "b ot",
+ "▁pub lish",
+ "▁publi sh",
+ "▁ publish",
+ ")* *",
+ ") **",
+ "AT TR",
+ "ATT R",
+ "ir al",
+ "ira l",
+ "i ral",
+ "V T",
+ "▁recogn ized",
+ "▁recognize d",
+ "▁L ind",
+ "▁Lin d",
+ "▁Li nd",
+ "ect ion",
+ "e ction",
+ "▁rel atively",
+ "▁relative ly",
+ "▁relativ ely",
+ "▁A h",
+ "▁ Ah",
+ "▁D ig",
+ "▁Di g",
+ "▁ Dig",
+ "ц ь",
+ "ic ket",
+ "ick et",
+ "▁specific ally",
+ "no st",
+ "nos t",
+ "n ost",
+ "▁g rass",
+ "▁gr ass",
+ "▁gra ss",
+ "▁gras s",
+ "▁c auses",
+ "▁caus es",
+ "▁cause s",
+ "▁ca uses",
+ "т во",
+ "ut ter",
+ "utt er",
+ "▁F estival",
+ "▁Fest ival",
+ "gr eg",
+ "gre g",
+ "g reg",
+ "▁weap ons",
+ "▁weapon s",
+ "▁s ir",
+ "▁si r",
+ "▁Virgin ia",
+ "lo gin",
+ "log in",
+ "▁s chedul",
+ "▁sched ul",
+ "сь кого",
+ "сько го",
+ "▁l osing",
+ "▁lo sing",
+ "▁los ing",
+ "▁E urop",
+ "▁Euro p",
+ "▁Eu rop",
+ "\"> <",
+ "\" ><",
+ "as p",
+ "a sp",
+ "aj o",
+ "a jo",
+ "ex ports",
+ "exp orts",
+ "export s",
+ "▁N ode",
+ "▁No de",
+ "▁ Node",
+ "▁j ako",
+ "▁ja ko",
+ "▁jak o",
+ "▁y a",
+ "▁ ya",
+ "▁success fully",
+ "▁successful ly",
+ "▁friend ly",
+ "▁ friendly",
+ "buf f",
+ "bu ff",
+ "b uff",
+ "DE FAULT",
+ "▁pre gn",
+ "▁preg n",
+ "Requ ired",
+ "Require d",
+ "▁b inary",
+ "▁bin ary",
+ "▁ binary",
+ "is ting",
+ "ist ing",
+ "isti ng",
+ "▁st ared",
+ "▁star ed",
+ "▁stare d",
+ "▁sta red",
+ "▁circum stances",
+ "▁х о",
+ "▁ хо",
+ "re i",
+ "r ei",
+ "▁Г о",
+ "Trans form",
+ "cn t",
+ "c nt",
+ "▁E xt",
+ "▁Ex t",
+ "▁ Ext",
+ "re port",
+ "rep ort",
+ "repo rt",
+ "VER SION",
+ "▁an aly",
+ "▁anal y",
+ "▁ analy",
+ "▁M arg",
+ "▁Mar g",
+ "▁Ma rg",
+ "▁al leg",
+ "▁all eg",
+ "▁alle g",
+ "build er",
+ "b uilder",
+ "To String",
+ "La yer",
+ "L ayer",
+ "ís t",
+ "í st",
+ "Pro p",
+ "Pr op",
+ "P rop",
+ "▁E mp",
+ "▁Em p",
+ "▁ Emp",
+ "} ]",
+ "▁s elling",
+ "▁sell ing",
+ "▁sel ling",
+ "▁ selling",
+ "▁que ue",
+ "▁ queue",
+ "▁ser iously",
+ "▁serious ly",
+ "▁L ead",
+ "▁Le ad",
+ "▁ Lead",
+ "text it",
+ "tex tit",
+ "test ing",
+ "tes ting",
+ "▁П ре",
+ "se curity",
+ "sec urity",
+ "ia ł",
+ "i ał",
+ "ú n",
+ "ch ip",
+ "chi p",
+ "c hip",
+ "▁c andidate",
+ "▁candid ate",
+ "▁min ister",
+ "▁mini ster",
+ "▁minist er",
+ "▁ minister",
+ "er ia",
+ "eri a",
+ "e ria",
+ "▁H et",
+ "▁He t",
+ "ди н",
+ "д ин",
+ "▁Brit ain",
+ "▁b arely",
+ "▁bar ely",
+ "▁bare ly",
+ "▁s ty",
+ "▁st y",
+ "▁ sty",
+ "▁Span ish",
+ "▁V en",
+ "▁Ve n",
+ "time r",
+ "ti mer",
+ "tim er",
+ "t imer",
+ "кі в",
+ "к ів",
+ "▁document s",
+ "▁doc uments",
+ "(' .",
+ "( '.",
+ "▁d ebug",
+ "▁de bug",
+ "▁deb ug",
+ "▁ debug",
+ "▁cont ro",
+ "▁contr o",
+ "сто я",
+ "▁j oy",
+ "▁jo y",
+ "▁ joy",
+ "S n",
+ "In v",
+ "I nv",
+ "▁pro tocol",
+ "▁proto col",
+ "▁prot ocol",
+ "▁ protocol",
+ "▁f aces",
+ "▁face s",
+ "▁fac es",
+ "▁fa ces",
+ "▁ faces",
+ "▁Des pite",
+ "se d",
+ "s ed",
+ "Con f",
+ "Co nf",
+ "AR G",
+ "A RG",
+ "▁e volution",
+ "▁ev olution",
+ "▁t od",
+ "▁to d",
+ "▁P romise",
+ "▁Prom ise",
+ "▁ Promise",
+ "▁pos ted",
+ "▁po sted",
+ "▁post ed",
+ "Per m",
+ "Pe rm",
+ "P erm",
+ "be t",
+ "b et",
+ "An g",
+ "A ng",
+ "J ust",
+ "▁r um",
+ "▁ru m",
+ "▁ rum",
+ "la yer",
+ "lay er",
+ "l ayer",
+ "▁beh avi",
+ "▁behav i",
+ "ip ping",
+ "ipp ing",
+ "ippi ng",
+ "i pping",
+ "▁d ynam",
+ "▁dy nam",
+ "▁dyn am",
+ "▁sch eme",
+ "▁sche me",
+ "▁ scheme",
+ "▁pro to",
+ "▁pr oto",
+ "▁prot o",
+ "▁ proto",
+ ") /",
+ "Col lections",
+ "Collection s",
+ "Collect ions",
+ "ri ev",
+ "rie v",
+ "r iev",
+ "▁C lick",
+ "▁Cl ick",
+ "▁ Click",
+ "▁u ns",
+ "▁un s",
+ "▁ uns",
+ "wide tilde",
+ "widet ilde",
+ "▁remember ed",
+ "г і",
+ "in ates",
+ "ina tes",
+ "inate s",
+ "▁incor por",
+ "▁De scription",
+ "▁Des cription",
+ "▁ Description",
+ "▁pre pare",
+ "▁prep are",
+ "▁prepar e",
+ "▁ prepare",
+ "▁F inal",
+ "▁Fin al",
+ "▁Fi nal",
+ "▁ Final",
+ "u ation",
+ "▁Qu een",
+ "▁Que en",
+ "> ;",
+ "▁autom atically",
+ "▁automatic ally",
+ "▁sh arp",
+ "▁shar p",
+ "▁sha rp",
+ "▁me at",
+ "at eur",
+ "ate ur",
+ "as tern",
+ "ast ern",
+ "aster n",
+ "aste rn",
+ "▁st uck",
+ "ASS ERT",
+ "▁pl anned",
+ "▁plan ned",
+ "do ts",
+ "dot s",
+ "d ots",
+ "ook ie",
+ "oo kie",
+ "▁His tor",
+ "▁Hist or",
+ "▁re views",
+ "▁review s",
+ "IM P",
+ "I MP",
+ "▁answ ered",
+ "▁answer ed",
+ "To tal",
+ "T otal",
+ "▁s au",
+ "▁sa u",
+ "▁Me xico",
+ "▁Mex ico",
+ "contin ue",
+ "▁App le",
+ "▁Ap ple",
+ "like ly",
+ "lik ely",
+ "з ва",
+ "us ers",
+ "use rs",
+ "user s",
+ "▁ident ified",
+ "▁L ev",
+ "▁Le v",
+ "▁m ol",
+ "▁mo l",
+ "▁Is lam",
+ "▁com mitted",
+ "▁comm itted",
+ "▁commit ted",
+ "wr it",
+ "w rit",
+ "бе р",
+ "б ер",
+ "ri ft",
+ "rif t",
+ "r ift",
+ "▁inter rupt",
+ "▁ interrupt",
+ "▁read only",
+ "sch ema",
+ "sche ma",
+ "s chema",
+ "S m",
+ "D ouble",
+ "az a",
+ "a za",
+ "▁H al",
+ "▁Ha l",
+ "▁ Hal",
+ "Mo ve",
+ "M ove",
+ "▁S eries",
+ "▁Se ries",
+ "▁Ser ies",
+ "▁Serie s",
+ "▁ Series",
+ "in line",
+ "▁кото ры",
+ "so c",
+ "s oc",
+ "▁t ent",
+ "▁te nt",
+ "▁ten t",
+ "▁a mer",
+ "▁am er",
+ "▁ amer",
+ "ak i",
+ "a ki",
+ "▁l ady",
+ "▁la dy",
+ "▁lad y",
+ "▁t ired",
+ "▁ti red",
+ "▁tire d",
+ "▁tir ed",
+ "if i",
+ "i fi",
+ "▁m ême",
+ "▁ même",
+ "ou ver",
+ "▁a side",
+ "▁as ide",
+ "Di d",
+ "D id",
+ "', \r",
+ "' ,\r",
+ "▁br inging",
+ "▁bring ing",
+ "Draw ing",
+ "ar o",
+ "a ro",
+ "▁R h",
+ "▁N az",
+ "▁Na z",
+ "es so",
+ "ess o",
+ "▁re action",
+ "▁react ion",
+ "mit ted",
+ "mitt ed",
+ "m itted",
+ "▁abs olute",
+ "▁absolut e",
+ "▁ absolute",
+ "ha ust",
+ "haus t",
+ "(( )",
+ "( ()",
+ "▁T ask",
+ "▁Ta sk",
+ "▁ Task",
+ "ER S",
+ "E RS",
+ "▁^ {",
+ "▁ ^{",
+ "V D",
+ "▁t one",
+ "▁to ne",
+ "▁ton e",
+ "dis t",
+ "di st",
+ "d ist",
+ "v s",
+ "▁whe el",
+ "▁ wheel",
+ "▁administr ation",
+ "▁admin istration",
+ "▁inter ests",
+ "▁interest s",
+ "▁point er",
+ "▁po inter",
+ "▁ pointer",
+ "▁en counter",
+ "▁enc ounter",
+ "ave r",
+ "av er",
+ "a ver",
+ "▁n ord",
+ "▁no rd",
+ "▁nor d",
+ "ke t",
+ "k et",
+ "▁b each",
+ "▁be ach",
+ "▁enjoy ed",
+ "cont ains",
+ "▁app end",
+ "▁ap pend",
+ "▁appe nd",
+ "▁ append",
+ "W ait",
+ "▁s quad",
+ "▁squ ad",
+ "ze l",
+ "z el",
+ "▁med ium",
+ "▁medi um",
+ "▁ medium",
+ "▁s ending",
+ "▁send ing",
+ "▁sen ding",
+ "▁L ady",
+ "▁La dy",
+ "▁Lad y",
+ "ç ões",
+ "▁dest ination",
+ "▁destin ation",
+ "▁ destination",
+ "ny ch",
+ "n ych",
+ "▁conf lict",
+ "▁conflic t",
+ "▁L y",
+ "▁v ul",
+ "▁vu l",
+ "▁bas ically",
+ "▁basic ally",
+ "re ated",
+ "reat ed",
+ "reate d",
+ "rea ted",
+ "bl ack",
+ "ug ins",
+ "ugin s",
+ "▁cal m",
+ "▁ca lm",
+ "ér ie",
+ "éri e",
+ "é rie",
+ "ha r",
+ "h ar",
+ "ла н",
+ "л ан",
+ "▁С е",
+ "w atch",
+ "▁P ut",
+ "▁Pu t",
+ "▁ Put",
+ "▁d ump",
+ "▁du mp",
+ "▁ dump",
+ "ac her",
+ "ach er",
+ "ache r",
+ "a cher",
+ "sc roll",
+ "scr oll",
+ "▁cl aimed",
+ "▁claim ed",
+ "▁ claimed",
+ "▁Cont rol",
+ "▁ Control",
+ "▁bl ind",
+ "en ti",
+ "ent i",
+ "▁Ke ep",
+ "▁ Keep",
+ "▁Develop ment",
+ "im ages",
+ "image s",
+ "ima ges",
+ "imag es",
+ "▁t ough",
+ "▁to ugh",
+ "▁tou gh",
+ "ge bra",
+ "geb ra",
+ "▁se pt",
+ "▁sep t",
+ "he w",
+ "h ew",
+ "▁s kill",
+ "▁sk ill",
+ "▁ski ll",
+ "▁ skill",
+ "▁T ay",
+ "▁Ta y",
+ "▁k tó",
+ "ow ner",
+ "own er",
+ "par e",
+ "pa re",
+ "p are",
+ "▁f ee",
+ "▁fe e",
+ "▁ fee",
+ "▁contin ues",
+ "▁continue s",
+ "▁continu es",
+ "▁k an",
+ "▁ka n",
+ "▁ kan",
+ "be s",
+ "b es",
+ "▁c ha",
+ "▁ch a",
+ "▁ cha",
+ "ov o",
+ "o vo",
+ "▁N ight",
+ "▁Ni ght",
+ "ict ure",
+ "sh ire",
+ "s hire",
+ "▁es say",
+ "▁ess ay",
+ "▁sup pose",
+ "▁supp ose",
+ "et ic",
+ "eti c",
+ "Ar t",
+ "A rt",
+ "ac on",
+ "aco n",
+ "a con",
+ "ll a",
+ "l la",
+ "word s",
+ "wor ds",
+ "w ords",
+ "▁compar ison",
+ "▁B E",
+ "▁ BE",
+ "▁challeng es",
+ "▁challenge s",
+ "▁o l",
+ "▁ ol",
+ "cite p",
+ "cit ep",
+ "▁F oot",
+ "▁Fo ot",
+ "▁ Foot",
+ "▁S uch",
+ "▁Su ch",
+ "▁ Such",
+ "▁p apers",
+ "▁paper s",
+ "▁pa pers",
+ "▁pap ers",
+ "act iv",
+ "qu er",
+ "que r",
+ "q uer",
+ "т я",
+ "▁Т о",
+ "сь кий",
+ "th ur",
+ "do ne",
+ "don e",
+ "d one",
+ "▁sh ock",
+ "▁ded icated",
+ "▁dedic ated",
+ "▁cor respond",
+ "▁correspon d",
+ "Se cond",
+ "Sec ond",
+ "▁b ull",
+ "▁bu ll",
+ "▁bul l",
+ "li fe",
+ "lif e",
+ "l ife",
+ "ind ent",
+ "inde nt",
+ "inden t",
+ "▁fig ures",
+ "▁figure s",
+ "▁And rew",
+ "▁Andre w",
+ "▁Andr ew",
+ "is p",
+ "i sp",
+ "▁fav our",
+ "зд а",
+ "з да",
+ "▁E lect",
+ "▁El ect",
+ "▁Ele ct",
+ "F ull",
+ "▁near by",
+ "▁Reg ister",
+ "▁ Register",
+ "Sc ale",
+ "Scal e",
+ "ic ations",
+ "ication s",
+ "и н",
+ "▁A M",
+ "▁ AM",
+ "pa ir",
+ "p air",
+ "▁pers pective",
+ "▁n os",
+ "▁no s",
+ "▁ nos",
+ "ap a",
+ "a pa",
+ "ost ał",
+ "osta ł",
+ "▁P ers",
+ "▁Per s",
+ "▁Pe rs",
+ "▁ Pers",
+ "ic er",
+ "ice r",
+ "i cer",
+ "▁pl astic",
+ "до в",
+ "д ов",
+ "ci ples",
+ "cipl es",
+ "cip les",
+ "z ą",
+ "cl os",
+ "c los",
+ "▁у ча",
+ "▁ Á",
+ "pl ugin",
+ "plug in",
+ "▁an gle",
+ "▁ang le",
+ "▁angl e",
+ "▁ angle",
+ "▁com mission",
+ "▁comm ission",
+ "▁fun ds",
+ "▁fund s",
+ "▁in du",
+ "▁ind u",
+ "▁d rawn",
+ "▁dr awn",
+ "▁draw n",
+ "á m",
+ "▁develop ing",
+ "▁seg ment",
+ "▁ segment",
+ "is me",
+ "ism e",
+ "sc r",
+ "s cr",
+ "▁l ies",
+ "▁li es",
+ "▁lie s",
+ "▁I L",
+ "▁ IL",
+ "▁a pi",
+ "▁ap i",
+ "▁ api",
+ "Ext ension",
+ "▁s cal",
+ "▁sc al",
+ "▁ scal",
+ "inst all",
+ "▁We ek",
+ "▁ Week",
+ "▁gen tle",
+ "▁gent le",
+ "▁Canad ian",
+ "▁d ialog",
+ "▁dial og",
+ "▁dia log",
+ "▁ dialog",
+ "▁art icles",
+ "▁article s",
+ "▁artic les",
+ "The me",
+ "Th eme",
+ "S M",
+ "▁B ul",
+ "▁Bu l",
+ "▁ Bul",
+ "▁l eur",
+ "▁le ur",
+ "▁s tom",
+ "▁st om",
+ "▁sto m",
+ "Pl ugin",
+ "▁по сле",
+ "▁пос ле",
+ "▁st ead",
+ "▁ste ad",
+ "▁ stead",
+ "▁ ś",
+ "ip her",
+ "iph er",
+ "i pher",
+ "▁pr ze",
+ "▁prz e",
+ "▁d raft",
+ "▁dr aft",
+ "▁ draft",
+ "bot tom",
+ "b ottom",
+ "▁{ };",
+ "▁{} ;",
+ "▁stay ed",
+ "fe ature",
+ "feat ure",
+ "▁v ot",
+ "▁vo t",
+ "▁fab ric",
+ "ç a",
+ "(' #",
+ "re a",
+ "r ea",
+ "▁re put",
+ "▁rep ut",
+ "▁C ir",
+ "▁Ci r",
+ "▁ Cir",
+ "▁A L",
+ "▁ AL",
+ "▁assert Equals",
+ "▁ assertEquals",
+ "result s",
+ "▁C ross",
+ "▁Cr oss",
+ "▁Cro ss",
+ "▁ Cross",
+ "urs day",
+ "▁a udio",
+ "▁aud io",
+ "▁ audio",
+ "▁g ap",
+ "▁ga p",
+ "▁stre ets",
+ "▁street s",
+ "▁scient ific",
+ "pl atform",
+ "▁a uss",
+ "▁au ss",
+ "▁aus s",
+ "▁C ro",
+ "▁Cr o",
+ "▁part ial",
+ "▁parti al",
+ "▁ partial",
+ "un c",
+ "u nc",
+ "▁cho ices",
+ "▁choice s",
+ "▁и ли",
+ "pr ed",
+ "pre d",
+ "p red",
+ "▁he ads",
+ "▁head s",
+ "▁ heads",
+ "ter day",
+ "▁N ick",
+ "▁Nic k",
+ "▁Ni ck",
+ "▁we ird",
+ "as ant",
+ "asa nt",
+ "▁represent ed",
+ "▁п и",
+ "▁ пи",
+ "D P",
+ "or ders",
+ "ord ers",
+ "order s",
+ "cl ock",
+ "c lock",
+ "▁H o",
+ "ar ters",
+ "art ers",
+ "arter s",
+ "arte rs",
+ "C md",
+ "og a",
+ "o ga",
+ "Key s",
+ "Ke ys",
+ "Re port",
+ "Rep ort",
+ "Repo rt",
+ "▁V ill",
+ "▁Vi ll",
+ "▁Vil l",
+ "▁M u",
+ "▁ Mu",
+ "▁own ed",
+ "▁ owned",
+ "SU CCESS",
+ "▁type of",
+ "▁ typeof",
+ "hd r",
+ "h dr",
+ "ua ble",
+ "u able",
+ "▁neighbor hood",
+ "▁A P",
+ "▁ AP",
+ "▁result ing",
+ "▁sh adow",
+ "▁ shadow",
+ "STR ING",
+ "▁video s",
+ "▁vide os",
+ "ле ння",
+ "лен ня",
+ "ex pect",
+ "exp ect",
+ "▁Val ley",
+ "▁Vall ey",
+ "▁g oto",
+ "▁go to",
+ "▁got o",
+ "▁ goto",
+ "▁S her",
+ "▁She r",
+ "▁Sh er",
+ "fr astr",
+ "▁oper ating",
+ "▁opera ting",
+ "▁э то",
+ "▁License d",
+ "▁Lic ensed",
+ "Var iable",
+ "Vari able",
+ "▁P R",
+ "▁ PR",
+ "▁H ans",
+ "▁Ha ns",
+ "▁Han s",
+ "cl one",
+ "▁G esch",
+ "▁Ge sch",
+ "▁Ges ch",
+ "▁B and",
+ "▁Ba nd",
+ "▁Ban d",
+ "▁ Band",
+ "... .....",
+ ".... ....",
+ "..... ...",
+ "ui ng",
+ "u ing",
+ "▁hundred s",
+ "▁о к",
+ "▁emot ional",
+ "▁emotion al",
+ "▁Ind ust",
+ ") +",
+ "▁Egy pt",
+ "▁fr anç",
+ "▁ š",
+ "▁f asc",
+ "▁fa sc",
+ "on to",
+ "ont o",
+ "▁A dam",
+ "▁Ad am",
+ "▁l aid",
+ "▁la id",
+ "▁r ig",
+ "▁ri g",
+ "▁ rig",
+ "▁det ailed",
+ "▁detail ed",
+ "▁im plements",
+ "▁implement s",
+ "▁impl ements",
+ "▁univers ity",
+ "▁H y",
+ "▁ Hy",
+ "▁g rid",
+ "▁gr id",
+ "▁gri d",
+ "▁ grid",
+ "▁reg ions",
+ "▁region s",
+ "St op",
+ "S top",
+ "▁s lot",
+ "▁sl ot",
+ "▁ slot",
+ "▁ang ry",
+ "▁- =",
+ "▁wait ed",
+ "▁wa ited",
+ "Ver t",
+ "V ert",
+ "\": \"",
+ "\" :\"",
+ "▁e lem",
+ "▁el em",
+ "▁ele m",
+ "▁ elem",
+ "▁r ég",
+ "▁ré g",
+ "ow ed",
+ "owe d",
+ "o wed",
+ "Mem ber",
+ "Me mber",
+ "M ember",
+ "▁r atio",
+ "▁rat io",
+ "▁ ratio",
+ "is en",
+ "ise n",
+ "i sen",
+ "▁L em",
+ "▁Le m",
+ "ge ry",
+ "ger y",
+ "g ery",
+ "▁c ream",
+ "▁cre am",
+ "▁ét ait",
+ "▁ était",
+ "▁g eb",
+ "▁ge b",
+ "▁ geb",
+ "un ique",
+ "uni que",
+ "▁D eb",
+ "▁De b",
+ "▁f actory",
+ "▁fact ory",
+ "▁factor y",
+ "▁ factory",
+ "ż e",
+ "d ialog",
+ "▁Con fig",
+ "▁Conf ig",
+ "▁ Config",
+ "Sy nc",
+ "S ync",
+ "an gers",
+ "ang ers",
+ "ange rs",
+ "anger s",
+ "▁gover ning",
+ "▁govern ing",
+ "▁H un",
+ "▁Hu n",
+ "Sp ace",
+ "S pace",
+ "▁j est",
+ "▁je st",
+ "ic ious",
+ "ici ous",
+ "icio us",
+ "▁em phas",
+ "▁emp has",
+ "um ps",
+ "ump s",
+ "▁E sp",
+ "▁Es p",
+ "▁ Esp",
+ "▁s ul",
+ "▁su l",
+ "▁histor ical",
+ "▁historic al",
+ "ij a",
+ "i ja",
+ "▁l ying",
+ "▁ly ing",
+ "▁ lying",
+ "▁St eve",
+ "▁Ste ve",
+ "▁me asures",
+ "▁measure s",
+ "▁meas ures",
+ "os to",
+ "ost o",
+ "o sto",
+ "? ”",
+ "▁p ocket",
+ "▁poc ket",
+ "▁S at",
+ "▁Sa t",
+ "▁p itch",
+ "▁pit ch",
+ "▁n atur",
+ "▁nat ur",
+ "▁hum ans",
+ "▁human s",
+ "▁Sim on",
+ "▁Si mon",
+ "ad ores",
+ "ado res",
+ "ador es",
+ "(\" \\",
+ "( \"\\",
+ "in king",
+ "ink ing",
+ "▁ex pos",
+ "▁exp os",
+ "mat erial",
+ "mate rial",
+ "m aterial",
+ "▁app arently",
+ "▁apparent ly",
+ "▁appar ently",
+ "▁C amb",
+ "▁Cam b",
+ "▁Ca mb",
+ "▁B ox",
+ "▁Bo x",
+ "▁ Box",
+ "▁s paces",
+ "▁sp aces",
+ "▁space s",
+ "ex ists",
+ "exist s",
+ "▁act ing",
+ "▁ac ting",
+ "OR Y",
+ "зо ва",
+ "Go od",
+ "G ood",
+ "ien ne",
+ "i enne",
+ "▁William s",
+ "▁f ruit",
+ "▁fr uit",
+ "▁fru it",
+ "ie ra",
+ "ier a",
+ "i era",
+ "▁L im",
+ "▁Li m",
+ "▁ Lim",
+ "▁t rait",
+ "▁tr ait",
+ "▁tra it",
+ "▁ trait",
+ "▁art ists",
+ "▁artist s",
+ "▁ab sor",
+ "▁abs or",
+ "ra it",
+ "rai t",
+ "r ait",
+ "LO AD",
+ "▁mov ies",
+ "▁movie s",
+ "▁d ynamic",
+ "▁dynam ic",
+ "▁dyn amic",
+ "▁ dynamic",
+ "as ts",
+ "ast s",
+ "a sts",
+ "▁In teger",
+ "▁ Integer",
+ "▁sm oke",
+ "п і",
+ "an gel",
+ "ang el",
+ "ange l",
+ ">( \"",
+ "> (\"",
+ "▁in strument",
+ "▁instr ument",
+ "▁f uel",
+ "▁fue l",
+ "▁fu el",
+ "но ї",
+ "atal ogue",
+ "atalog ue",
+ "▁s erial",
+ "▁se rial",
+ "▁ser ial",
+ "▁ serial",
+ "File s",
+ "Fil es",
+ "Fi les",
+ "F iles",
+ "▁bath room",
+ "il o",
+ "i lo",
+ "es to",
+ "est o",
+ "e sto",
+ "▁p m",
+ "▁ pm",
+ "ent ials",
+ "ential s",
+ "enti als",
+ "▁On line",
+ "wh ite",
+ "▁t ips",
+ "▁tip s",
+ "▁ti ps",
+ "▁cap able",
+ "Fi g",
+ "F ig",
+ "T V",
+ "▁о н",
+ "▁ он",
+ "k é",
+ "bit r",
+ "bi tr",
+ "b itr",
+ "Map ping",
+ "Ma pping",
+ "M apping",
+ "▁t ak",
+ "▁ta k",
+ "ю щи",
+ "в ля",
+ ")\" ,",
+ ") \",",
+ "▁K arl",
+ "▁Kar l",
+ "▁Ka rl",
+ "▁H uman",
+ "▁Hu man",
+ "▁Hum an",
+ "▁P ot",
+ "▁Po t",
+ "▁rep resents",
+ "▁represent s",
+ "▁cons istent",
+ "▁consist ent",
+ "_ (",
+ "we n",
+ "w en",
+ "▁R ose",
+ "▁Ro se",
+ "▁Ros e",
+ "la w",
+ "l aw",
+ "▁F ROM",
+ "▁FR OM",
+ "▁ FROM",
+ "▁beg ins",
+ "▁begin s",
+ "▁e dit",
+ "▁ed it",
+ "▁ edit",
+ "▁mount ain",
+ "▁ch apter",
+ "▁chap ter",
+ "▁wonder ed",
+ "▁indust rial",
+ "▁M ajor",
+ "▁Ma jor",
+ "▁Maj or",
+ "▁g es",
+ "▁ge s",
+ "▁ ges",
+ "▁direct ed",
+ "▁dire cted",
+ "er os",
+ "ero s",
+ "e ros",
+ "▁W ild",
+ "▁Wil d",
+ "▁Wi ld",
+ "li ament",
+ "lia ment",
+ "Bo ok",
+ "B ook",
+ "user name",
+ "ho t",
+ "h ot",
+ "▁n am",
+ "▁na m",
+ "▁ nam",
+ "▁le ague",
+ "br a",
+ "b ra",
+ "ко н",
+ "к он",
+ "▁T al",
+ "▁Ta l",
+ "▁В а",
+ "▁ex ports",
+ "▁exp orts",
+ "▁export s",
+ "▁ exports",
+ "( @",
+ "▁sh aring",
+ "▁shar ing",
+ "▁sha ring",
+ "▁T ro",
+ "▁Tr o",
+ "ś ć",
+ "ues day",
+ "yl v",
+ "y lv",
+ "▁gu itar",
+ "el en",
+ "ele n",
+ "e len",
+ "Se lection",
+ "Select ion",
+ "S election",
+ "▁conf ident",
+ "ry pto",
+ "rypt o",
+ "▁h ors",
+ "▁hor s",
+ "▁ho rs",
+ "ed itor",
+ "edit or",
+ "edi tor",
+ "▁should ers",
+ "▁shoulder s",
+ "get Name",
+ "en cing",
+ "enc ing",
+ "enci ng",
+ "SE LECT",
+ "SEL ECT",
+ "в ши",
+ "▁kind s",
+ "▁kin ds",
+ "▁W el",
+ "▁We l",
+ "▁pur poses",
+ "▁purpose s",
+ "Mat rix",
+ "in valid",
+ "▁own ers",
+ "▁owner s",
+ "▁ owners",
+ "▁Rec ords",
+ "▁Record s",
+ "▁ Records",
+ "▁Pro cess",
+ "▁ Process",
+ "▁c hat",
+ "▁ch at",
+ "▁cha t",
+ "▁ chat",
+ "▁D or",
+ "▁Do r",
+ "▁b in",
+ "▁bi n",
+ "▁ bin",
+ "re dit",
+ "red it",
+ "r edit",
+ "oi re",
+ "oir e",
+ "o ire",
+ "▁T otal",
+ "▁To tal",
+ "▁Tot al",
+ "▁ Total",
+ "▁F amily",
+ "▁Famil y",
+ "▁ Family",
+ "AR Y",
+ "▁b read",
+ "▁br ead",
+ "▁bre ad",
+ "▁ bread",
+ "▁com pre",
+ "▁comp re",
+ "▁compr e",
+ "▁sh oes",
+ "▁shoe s",
+ "▁r az",
+ "▁ra z",
+ "▁ raz",
+ "▁tr ace",
+ "▁tra ce",
+ "▁ trace",
+ "ne j",
+ "n ej",
+ "or ted",
+ "ort ed",
+ "orte d",
+ "h n",
+ "▁pro cedure",
+ "▁proced ure",
+ "pro perties",
+ "pl ier",
+ "▁h ero",
+ "▁he ro",
+ "▁her o",
+ "▁ hero",
+ "pan el",
+ "pa nel",
+ "p anel",
+ "▁mark ed",
+ "▁mar ked",
+ "▁wor ried",
+ "\\ |",
+ "pt s",
+ "p ts",
+ "▁S upport",
+ "▁Sup port",
+ "▁Supp ort",
+ "▁ Support",
+ "▁ser ving",
+ "▁serv ing",
+ "F ail",
+ "▁dis appoint",
+ "▁Sc ot",
+ "▁ple asure",
+ "▁j udge",
+ "▁jud ge",
+ "▁judg e",
+ "ze ich",
+ "▁for ever",
+ "▁fore ver",
+ "▁Ze it",
+ "uo us",
+ "u ous",
+ "in ent",
+ "ine nt",
+ "inen t",
+ "i nent",
+ "▁d w",
+ "▁ dw",
+ "▁w aren",
+ "▁war en",
+ "▁wa ren",
+ "▁ware n",
+ "▁fl ash",
+ "▁ flash",
+ "▁tro ops",
+ "▁dr ugs",
+ "▁dru gs",
+ "▁drug s",
+ "▁d iam",
+ "▁di am",
+ "▁dia m",
+ ". ~",
+ "im p",
+ "i mp",
+ "in ned",
+ "inn ed",
+ "▁E V",
+ "▁ EV",
+ "St ruct",
+ "Str uct",
+ "▁just ice",
+ "▁offic ials",
+ "▁official s",
+ "ff ff",
+ "fff f",
+ "f fff",
+ "▁Com mon",
+ "▁Comm on",
+ "▁ Common",
+ "▁C at",
+ "▁Ca t",
+ "▁ Cat",
+ "▁tom orrow",
+ "▁é l",
+ "▁ él",
+ "Text ure",
+ "Te xture",
+ "qp oint",
+ "q point",
+ "▁F ried",
+ "▁Fr ied",
+ "▁T erm",
+ "▁Te rm",
+ "▁Ter m",
+ "▁ Term",
+ "pgf qpoint",
+ "▁n em",
+ "▁ne m",
+ "▁ nem",
+ "no rm",
+ "nor m",
+ "n orm",
+ "▁hard ly",
+ "od a",
+ "o da",
+ "ze ta",
+ "zet a",
+ "z eta",
+ "em ic",
+ "emi c",
+ "e mic",
+ "▁по лу",
+ "▁пол у",
+ "▁lo aded",
+ "▁load ed",
+ "▁ loaded",
+ "ke s",
+ "k es",
+ "ci ó",
+ "c ió",
+ "▁f ool",
+ "▁fo ol",
+ "▁foo l",
+ "▁t rick",
+ "▁tr ick",
+ "▁tri ck",
+ "▁d st",
+ "▁ds t",
+ "▁ dst",
+ "Fin d",
+ "Fi nd",
+ "F ind",
+ "▁в се",
+ "}} ,",
+ "} },",
+ "▁frame work",
+ "▁ framework",
+ "▁mer ely",
+ "▁mere ly",
+ "▁un ion",
+ "▁ union",
+ "▁Ed ward",
+ "ri f",
+ "r if",
+ "Fl ag",
+ "F lag",
+ "▁cris is",
+ "▁cri sis",
+ "▁fin ite",
+ "▁ finite",
+ "▁l ol",
+ "▁lo l",
+ "▁K im",
+ "▁Ki m",
+ "на та",
+ "sin ce",
+ "s ince",
+ "▁com pat",
+ "▁comp at",
+ "▁ compat",
+ "▁p ert",
+ "▁per t",
+ "▁pe rt",
+ "▁ pert",
+ "ib ilities",
+ "ibil ities",
+ "▁tamb ién",
+ "ib li",
+ "▁t een",
+ "▁te en",
+ "▁ teen",
+ "▁sym pt",
+ "or al",
+ "ora l",
+ "o ral",
+ "de rs",
+ "der s",
+ "d ers",
+ "ot te",
+ "ott e",
+ "п ри",
+ "▁J ane",
+ "▁Jan e",
+ "▁Ja ne",
+ "▁original ly",
+ "▁origin ally",
+ "▁thro at",
+ "ma g",
+ "m ag",
+ "su p",
+ "s up",
+ "un i",
+ "u ni",
+ "$ $",
+ "▁L ibrary",
+ "▁ Library",
+ "▁att acks",
+ "▁attack s",
+ "in gen",
+ "ing en",
+ "inge n",
+ "(' /",
+ "▁h es",
+ "▁he s",
+ "▁ hes",
+ "co in",
+ "c oin",
+ "oun ce",
+ "▁Academ y",
+ "MOD ULE",
+ "is ms",
+ "ism s",
+ "▁A dv",
+ "▁Ad v",
+ "▁ Adv",
+ "▁B ol",
+ "▁Bo l",
+ "▁inc ident",
+ ")^ {",
+ ") ^{",
+ "▁b ij",
+ "▁bi j",
+ "▁R ome",
+ "▁Rom e",
+ "▁Ro me",
+ "▁It aly",
+ "▁Ital y",
+ "ev ents",
+ "event s",
+ "even ts",
+ "▁F ern",
+ "▁Fe rn",
+ "▁Fer n",
+ "▁b er",
+ "▁be r",
+ "▁ ber",
+ "▁sil ent",
+ "▁p ier",
+ "▁pie r",
+ "▁pi er",
+ "▁Y O",
+ "▁pl ain",
+ "▁ plain",
+ "B as",
+ "▁p ill",
+ "▁pi ll",
+ "▁pil l",
+ "ra se",
+ "ras e",
+ "r ase",
+ "▁car rying",
+ "▁carry ing",
+ "▁re sp",
+ "▁r esp",
+ "▁res p",
+ "▁ resp",
+ "ну ю",
+ "▁typ ical",
+ "Wrap per",
+ "W rapper",
+ "▁g au",
+ "▁ga u",
+ "▁chem ical",
+ "▁h al",
+ "▁ha l",
+ "▁ hal",
+ "th row",
+ "Cl uster",
+ "▁G ab",
+ "▁Ga b",
+ "▁G irl",
+ "▁Gi rl",
+ "▁Gir l",
+ "qu ir",
+ "▁A rg",
+ "▁Ar g",
+ "▁ Arg",
+ "▁rel ief",
+ "▁relie f",
+ "▁reli ef",
+ "▁В е",
+ "d m",
+ "▁fr ustr",
+ "▁fru str",
+ "\\ %",
+ "▁st ores",
+ "▁store s",
+ "▁stor es",
+ "▁sto res",
+ "▁bott le",
+ "▁bot tle",
+ "▁L ew",
+ "▁Le w",
+ "tw o",
+ "t wo",
+ "st ad",
+ "sta d",
+ "▁che ek",
+ "▁concern s",
+ "▁concer ns",
+ "▁help ful",
+ "▁co verage",
+ "▁cover age",
+ "is i",
+ "i si",
+ "AD D",
+ "A DD",
+ "as ync",
+ "asy nc",
+ "a sync",
+ "▁approxim ately",
+ "▁approx imately",
+ "▁approximate ly",
+ "if fer",
+ "iff er",
+ "iffe r",
+ "ho ok",
+ "h ook",
+ "▁e num",
+ "▁en um",
+ "▁ enum",
+ "ov á",
+ "o vá",
+ "▁e vil",
+ "▁ev il",
+ "▁const antly",
+ "▁constant ly",
+ "ap ply",
+ "app ly",
+ "▁si è",
+ "▁pract ices",
+ "▁practice s",
+ "▁te achers",
+ "▁teach ers",
+ "▁teacher s",
+ "▁S n",
+ "▁ Sn",
+ "▁A wards",
+ "▁Award s",
+ "▁Aw ards",
+ "▁sub stant",
+ "▁subst ant",
+ "▁$ .",
+ "▁ $.",
+ "d k",
+ "▁m ob",
+ "▁mo b",
+ "▁ mob",
+ "▁ing red",
+ "ve re",
+ "ver e",
+ "v ere",
+ "Mult i",
+ "пе р",
+ "п ер",
+ "st al",
+ "sta l",
+ "s tal",
+ "ya rd",
+ "yar d",
+ "y ard",
+ "requ ired",
+ "require d",
+ "ve ment",
+ "v ement",
+ "▁int elligence",
+ "▁intellig ence",
+ "▁th inks",
+ "▁think s",
+ "▁thin ks",
+ "▁person ally",
+ "▁personal ly",
+ "▁tr ained",
+ "▁tra ined",
+ "▁train ed",
+ "▁ trained",
+ "or ney",
+ "orn ey",
+ "orne y",
+ ") ",
+ "gg ed",
+ "g ged",
+ "E INVAL",
+ "ar na",
+ "arn a",
+ "▁Ham ilton",
+ "mer ce",
+ "ek t",
+ "e kt",
+ "O F",
+ ") [",
+ "ru g",
+ "r ug",
+ "ic ión",
+ "ici ón",
+ "ició n",
+ "i ción",
+ "▁sur vey",
+ "▁surv ey",
+ "▁surve y",
+ "nes day",
+ "▁p ag",
+ "▁pa g",
+ "▁ pag",
+ "▁bound ary",
+ "▁quant um",
+ "▁draw ing",
+ "▁vol unte",
+ "▁volunt e",
+ "▁W ord",
+ "▁Wo rd",
+ "▁Wor d",
+ "▁ Word",
+ "sk y",
+ "s ky",
+ "▁G reg",
+ "▁Gr eg",
+ "▁Gre g",
+ "co ll",
+ "col l",
+ "c oll",
+ "hi de",
+ "hid e",
+ "h ide",
+ "▁sw im",
+ "▁reve aled",
+ "▁reveal ed",
+ "ad v",
+ "a dv",
+ "д я",
+ ".\" );",
+ ".\") ;",
+ ". \");",
+ "▁ex plan",
+ "▁expl an",
+ "▁exp lan",
+ "▁Cur rent",
+ "▁ Current",
+ "▁got ten",
+ "▁f alling",
+ "▁fall ing",
+ "▁fal ling",
+ "▁cont ained",
+ "▁contain ed",
+ "UN D",
+ "U ND",
+ "▁Sh ould",
+ "▁ Should",
+ "▁k illing",
+ "▁kill ing",
+ "▁kil ling",
+ "▁aspect s",
+ "ic ted",
+ "ict ed",
+ "i cted",
+ "▁P aram",
+ "▁Par am",
+ "▁Pa ram",
+ "▁Para m",
+ "▁ Param",
+ "\", \r",
+ "\" ,\r",
+ "TI ON",
+ "T ION",
+ ")) ;\r",
+ ")); \r",
+ ") );\r",
+ "▁I ran",
+ "▁Ir an",
+ "▁Ira n",
+ "be it",
+ "▁B u",
+ "▁ Bu",
+ "▁[ ],",
+ "▁[] ,",
+ "▁ [],",
+ "SS ION",
+ "S SION",
+ "▁M ah",
+ "▁Ma h",
+ "▁res olution",
+ "▁b oss",
+ "▁bo ss",
+ "▁bos s",
+ "l g",
+ "ch or",
+ "cho r",
+ "c hor",
+ "▁Un ter",
+ "▁de bt",
+ "▁deb t",
+ "▁v id",
+ "▁vi d",
+ "▁ vid",
+ "gi e",
+ "g ie",
+ "▁u no",
+ "▁un o",
+ "▁ uno",
+ "C B",
+ "pl om",
+ "plo m",
+ "LIC ENSE",
+ "L ICENSE",
+ "▁K enn",
+ "▁Ke nn",
+ "▁Ken n",
+ "▁fin ns",
+ "ON G",
+ "O NG",
+ "▁some what",
+ "▁a ctor",
+ "▁act or",
+ "▁ac tor",
+ "▁ actor",
+ "▁St atus",
+ "▁Stat us",
+ "▁ Status",
+ "▁prob ability",
+ "f b",
+ "▁ch art",
+ "▁char t",
+ "▁cha rt",
+ "▁ chart",
+ "▁st ands",
+ "▁stand s",
+ "▁stan ds",
+ "pol icy",
+ "▁o nder",
+ "▁on der",
+ "▁onde r",
+ "▁ onder",
+ "tab ular",
+ "▁A sh",
+ "▁As h",
+ "▁bo ost",
+ "▁ boost",
+ "▁des per",
+ "▁desp er",
+ "mon th",
+ "mont h",
+ "▁al ert",
+ "▁ale rt",
+ "▁ alert",
+ "▁su ite",
+ "▁suit e",
+ "▁ suite",
+ "▁g én",
+ "▁gé n",
+ "▁v acc",
+ "▁va cc",
+ "▁vac c",
+ "▁H as",
+ "▁Ha s",
+ "▁ Has",
+ "Ma sk",
+ "M ask",
+ "▁Th ursday",
+ "▁pro ved",
+ "▁pr oved",
+ "▁prov ed",
+ "▁prove d",
+ "▁N el",
+ "▁Ne l",
+ "▁m oral",
+ "▁mor al",
+ "▁mo ral",
+ "▁j a",
+ "▁ ja",
+ "au er",
+ "a uer",
+ "co dec",
+ "code c",
+ "cod ec",
+ "▁in stant",
+ "▁inst ant",
+ "am ps",
+ "amp s",
+ "▁mil k",
+ "WO RD",
+ "W ORD",
+ "▁ Ö",
+ "Em ail",
+ "E mail",
+ "Element s",
+ "El ements",
+ "Elem ents",
+ "▁for ma",
+ "▁form a",
+ "Fr ee",
+ "F ree",
+ "MA P",
+ "M AP",
+ "▁ Ж",
+ "sy m",
+ "s ym",
+ "▁т и",
+ "▁ ти",
+ "▁E conom",
+ "▁Ec onom",
+ "▁V i",
+ "▁ Vi",
+ "▁Col umb",
+ "▁_ ,",
+ "▁ _,",
+ "or et",
+ "ore t",
+ "o ret",
+ "Se qu",
+ "Seq u",
+ "S equ",
+ "pl an",
+ "p lan",
+ "▁f requency",
+ "▁frequ ency",
+ "▁ frequency",
+ "ir ement",
+ "ire ment",
+ "▁ass umed",
+ "▁assum ed",
+ "▁assume d",
+ "▁C a",
+ "▁B it",
+ "▁Bi t",
+ "▁ Bit",
+ "▁ко ман",
+ "▁ком ан",
+ "▁sm ell",
+ "Se curity",
+ "Sec urity",
+ "▁a qu",
+ "▁ aqu",
+ "oo r",
+ "o or",
+ "pr ice",
+ "p rice",
+ "in ity",
+ "init y",
+ "ini ty",
+ "▁a xis",
+ "▁ax is",
+ "▁ axis",
+ "re lease",
+ "▁res olve",
+ "▁ resolve",
+ "▁t ears",
+ "▁te ars",
+ "▁tea rs",
+ "▁tear s",
+ "▁b other",
+ "▁bo ther",
+ "▁both er",
+ "▁bot her",
+ "▁Comm unity",
+ "▁Commun ity",
+ "▁register ed",
+ "▁re volution",
+ "▁rev olution",
+ "▁revol ution",
+ "? .",
+ "▁version s",
+ "▁vers ions",
+ "▁ versions",
+ "%% %%",
+ "yd ro",
+ "y dro",
+ "Su ccess",
+ "▁W in",
+ "▁Wi n",
+ "▁ Win",
+ "▁B oy",
+ "▁Bo y",
+ "▁D ub",
+ "▁Du b",
+ "▁k w",
+ "▁ kw",
+ "▁n och",
+ "▁no ch",
+ "▁char ges",
+ "▁charg es",
+ "▁charge s",
+ "ar ios",
+ "ari os",
+ "ario s",
+ "a rios",
+ "ua r",
+ "u ar",
+ "; &",
+ "▁hab ía",
+ "( `",
+ "▁t x",
+ "▁ tx",
+ "el ve",
+ "▁a ños",
+ "▁año s",
+ "▁m ath",
+ "▁mat h",
+ "▁ma th",
+ "▁ math",
+ "▁Al f",
+ "▁F und",
+ "▁Fun d",
+ "▁Fu nd",
+ "▁man ifest",
+ "▁manif est",
+ "▁att ached",
+ "▁attach ed",
+ "▁spirit ual",
+ "▁Alex ander",
+ "▁Alexand er",
+ "un es",
+ "une s",
+ "u nes",
+ "▁s eed",
+ "▁se ed",
+ "▁see d",
+ "▁ seed",
+ "▁Н о",
+ "▁mag azine",
+ "▁magaz ine",
+ "▁e igen",
+ "▁о бра",
+ "▁об ра",
+ "▁ обра",
+ "e a",
+ "▁P H",
+ "▁ PH",
+ "sw ing",
+ "s wing",
+ "▁As ia",
+ "ј у",
+ "▁K IND",
+ "Ident ifier",
+ "on ce",
+ "▁al cohol",
+ "ці ї",
+ "st yles",
+ "style s",
+ "sty les",
+ "assert Equal",
+ "▁R a",
+ "гра фи",
+ "▁mill ions",
+ "▁million s",
+ "▁ch unk",
+ "▁ chunk",
+ "де р",
+ "д ер",
+ "Pack age",
+ "US T",
+ "U ST",
+ "▁N othing",
+ "▁No thing",
+ "▁Not hing",
+ "▁ Nothing",
+ "(\" #",
+ "▁M id",
+ "▁Mi d",
+ "▁на ча",
+ "▁ нача",
+ "ł y",
+ "AA AA",
+ "▁la unched",
+ "▁launch ed",
+ "▁w ake",
+ "▁wa ke",
+ "▁ wake",
+ "▁gu ests",
+ "▁guest s",
+ "▁dif ferences",
+ "▁differ ences",
+ "▁difference s",
+ "ud i",
+ "u di",
+ "▁a id",
+ "▁ai d",
+ "▁ aid",
+ "▁S port",
+ "▁Sp ort",
+ "ul ator",
+ "ula tor",
+ "ex ecute",
+ "exec ute",
+ "execut e",
+ "pl ot",
+ "plo t",
+ "p lot",
+ "ch ing",
+ "chi ng",
+ "c hing",
+ "▁N orm",
+ "▁No rm",
+ "▁Nor m",
+ "▁ Norm",
+ "t m",
+ "\\ +",
+ "AR D",
+ "A RD",
+ "▁be er",
+ "▁п ід",
+ "▁пі д",
+ "IA L",
+ "I AL",
+ "st orage",
+ "sto rage",
+ "▁An na",
+ "▁Ann a",
+ "▁y ards",
+ "▁yard s",
+ "▁techn ique",
+ "▁o ù",
+ "at ten",
+ "att en",
+ "atte n",
+ "UN T",
+ "U NT",
+ "do n",
+ "d on",
+ "фо р",
+ "ф ор",
+ "▁h oping",
+ "▁hop ing",
+ "▁ho ping",
+ "▁vict ory",
+ "it at",
+ "ita t",
+ "i tat",
+ "▁signific antly",
+ "▁significant ly",
+ "▁pract ical",
+ "ij e",
+ "i je",
+ "▁exp ansion",
+ "▁expans ion",
+ "J S",
+ "ix els",
+ "ixel s",
+ "US ER",
+ "USE R",
+ "U SER",
+ "Sh ape",
+ "▁ext ent",
+ "li o",
+ "l io",
+ "▁p ued",
+ "▁pu ed",
+ "ol id",
+ "oli d",
+ "▁g am",
+ "▁ga m",
+ "▁s event",
+ "▁se vent",
+ "▁seven t",
+ "▁G a",
+ "▁ Ga",
+ "angu ages",
+ "anguage s",
+ "(( (",
+ "( ((",
+ "ъ л",
+ "▁Ex per",
+ "▁Exp er",
+ "▁ Exper",
+ "as ty",
+ "ast y",
+ "a sty",
+ "ri eg",
+ "rie g",
+ "r ieg",
+ "gi o",
+ "g io",
+ "od o",
+ "o do",
+ "▁col le",
+ "▁co lle",
+ "▁coll e",
+ "▁st ored",
+ "▁store d",
+ "▁stor ed",
+ "▁sto red",
+ "▁S che",
+ "▁Sch e",
+ "▁Sc he",
+ "▁ Sche",
+ "ist ant",
+ "ista nt",
+ "istan t",
+ "i stant",
+ "▁l ip",
+ "▁li p",
+ "B R",
+ "▁a ug",
+ "▁au g",
+ "▁ aug",
+ "▁S earch",
+ "▁Se arch",
+ "▁ Search",
+ ")= \\",
+ ") =\\",
+ "▁U r",
+ "▁s ole",
+ "▁so le",
+ "▁sol e",
+ "▁ sole",
+ "il lo",
+ "ill o",
+ "▁me hr",
+ "ki t",
+ "k it",
+ "▁in terior",
+ "▁inter ior",
+ "▁inte rior",
+ "LI ST",
+ "L IST",
+ "ad el",
+ "ade l",
+ "a del",
+ "▁shop ping",
+ "▁s lä",
+ "▁sl ä",
+ "You r",
+ "Y our",
+ "DI TION",
+ "D ITION",
+ "▁H ttp",
+ "▁ Http",
+ "ra ham",
+ "rah am",
+ "т ри",
+ "▁b rings",
+ "▁br ings",
+ "▁bring s",
+ "Re v",
+ "R ev",
+ "▁pro pag",
+ "▁prop ag",
+ "ity Engine",
+ "() ),",
+ "()) ,",
+ "( )),",
+ "▁ing år",
+ "▁Ir eland",
+ "▁Ire land",
+ "▁\" ./",
+ "▁\". /",
+ "▁H arr",
+ "▁Har r",
+ "▁Ha rr",
+ "▁ad min",
+ "▁adm in",
+ "▁ admin",
+ "en o",
+ "e no",
+ "▁k r",
+ "▁ kr",
+ "▁est á",
+ "▁pro ps",
+ "▁pr ops",
+ "▁prop s",
+ "▁ props",
+ "to k",
+ "t ok",
+ "om orph",
+ "▁affect ed",
+ "Ph one",
+ "▁deg rees",
+ "▁degree s",
+ "so me",
+ "som e",
+ "s ome",
+ "▁n in",
+ "▁ni n",
+ "EV ENT",
+ "▁inter action",
+ "▁inte raction",
+ "▁interact ion",
+ "▁T uesday",
+ "iter ator",
+ "▁N ob",
+ "▁No b",
+ "▁sc atter",
+ "uck et",
+ "uc ket",
+ "com plete",
+ "comp lete",
+ "▁d uty",
+ "▁du ty",
+ "▁dut y",
+ "▁answ ers",
+ "▁answer s",
+ "Pro gress",
+ "ee d",
+ "e ed",
+ "ро н",
+ "р он",
+ "▁v ie",
+ "▁vi e",
+ "▁de pos",
+ "▁dep os",
+ "▁p acket",
+ "▁pack et",
+ "▁pac ket",
+ "▁ packet",
+ "▁t ow",
+ "▁to w",
+ "▁de leg",
+ "▁del eg",
+ "▁ deleg",
+ "aud io",
+ "a udio",
+ "▁v ary",
+ "▁var y",
+ "▁va ry",
+ "▁m igr",
+ "▁mi gr",
+ "▁mig r",
+ "▁ migr",
+ "ф і",
+ "es a",
+ "e sa",
+ "Event s",
+ "Ev ents",
+ "Even ts",
+ "ha us",
+ "h aus",
+ "▁S av",
+ "▁Sa v",
+ "▁Port ug",
+ "▁с то",
+ "▁ст о",
+ "▁ сто",
+ "il ation",
+ "i lation",
+ "▁met adata",
+ "▁meta data",
+ "▁ metadata",
+ "la s",
+ "l as",
+ "▁a i",
+ "▁ ai",
+ "▁an ger",
+ "▁ang er",
+ "▁ange r",
+ "▁ anger",
+ "▁h am",
+ "▁ha m",
+ "▁ ham",
+ "▁A nal",
+ "▁An al",
+ "▁Ana l",
+ "▁ Anal",
+ "▁frequ ently",
+ "▁frequent ly",
+ "▁F ALSE",
+ "▁ FALSE",
+ "oc he",
+ "och e",
+ "o che",
+ "re z",
+ "r ez",
+ "▁V iet",
+ "▁Vi et",
+ "qu is",
+ "q uis",
+ "▁char ged",
+ "▁charg ed",
+ "▁charge d",
+ "ä s",
+ "▁P ath",
+ "▁Pat h",
+ "▁Pa th",
+ "▁ Path",
+ "▁accur ate",
+ "▁Pl us",
+ "▁ Plus",
+ "ke it",
+ "▁In put",
+ "▁ Input",
+ "wh en",
+ "whe n",
+ "w hen",
+ "er as",
+ "era s",
+ "e ras",
+ "▁во з",
+ "▁de rived",
+ "▁der ived",
+ "▁deriv ed",
+ "▁derive d",
+ "aj e",
+ "a je",
+ "▁H ad",
+ "▁Ha d",
+ "ur en",
+ "ure n",
+ "u ren",
+ "ó r",
+ "}= \\",
+ "} =\\",
+ "ur eau",
+ "ure au",
+ "al and",
+ "ala nd",
+ "a land",
+ "Execut ion",
+ "Exec ution",
+ "ed en",
+ "ede n",
+ "e den",
+ "▁se eking",
+ "▁see king",
+ "▁seek ing",
+ "ch anged",
+ "change d",
+ "chan ged",
+ "▁t rem",
+ "▁tr em",
+ "▁tre m",
+ "ск у",
+ "с ку",
+ "▁G eme",
+ "▁Ge me",
+ "▁Gem e",
+ "in ating",
+ "ina ting",
+ "▁column s",
+ "▁ columns",
+ "E P",
+ "▁inj ury",
+ "end ent",
+ "ende nt",
+ "enden t",
+ "▁he aded",
+ "▁head ed",
+ "▁ headed",
+ "AS E",
+ "A SE",
+ "▁Mus lim",
+ "▁cl imate",
+ "▁clim ate",
+ "▁f ake",
+ "▁fa ke",
+ "▁ fake",
+ "CM D",
+ "C MD",
+ "ј и",
+ "▁Ar ts",
+ "▁Art s",
+ "fe ction",
+ "fect ion",
+ "f ection",
+ "▁p it",
+ "▁pi t",
+ "▁ pit",
+ "> \\",
+ "an al",
+ "ana l",
+ "a nal",
+ "Se ction",
+ "S ection",
+ "pl us",
+ "ü t",
+ "▁em bed",
+ "▁emb ed",
+ "▁ embed",
+ "▁st rings",
+ "▁str ings",
+ "▁string s",
+ "▁ strings",
+ "Be fore",
+ "B efore",
+ "pro c",
+ "pr oc",
+ "p roc",
+ "▁с по",
+ "▁сп о",
+ "▁ спо",
+ "tr l",
+ "t rl",
+ "v r",
+ "Back ground",
+ "log ger",
+ "ag raph",
+ "agr aph",
+ "agra ph",
+ "a graph",
+ "ie st",
+ "ies t",
+ "i est",
+ "▁good s",
+ "bat ch",
+ "b atch",
+ "▁opt ional",
+ "▁option al",
+ "▁ optional",
+ "▁Tay lor",
+ "▁recogn ize",
+ "wal k",
+ "w alk",
+ "▁H it",
+ "▁Hi t",
+ "▁ Hit",
+ "▁Eliz abeth",
+ "} :",
+ "▁care ful",
+ "кра ї",
+ "▁loc ations",
+ "▁location s",
+ "▁struct ures",
+ "▁structure s",
+ "▁d isk",
+ "▁dis k",
+ "▁di sk",
+ "▁ disk",
+ "▁sh ips",
+ "▁ship s",
+ "▁ ships",
+ "▁s uo",
+ "▁su o",
+ "▁s owie",
+ "▁so wie",
+ "▁sow ie",
+ "▁E ss",
+ "▁Es s",
+ "▁H ash",
+ "▁Ha sh",
+ "▁Has h",
+ "▁ Hash",
+ "▁reason able",
+ "▁More over",
+ "▁form ula",
+ "▁C entre",
+ "▁Cent re",
+ "▁res idents",
+ "▁resident s",
+ "▁resid ents",
+ "R S",
+ "Id s",
+ "I ds",
+ "▁K now",
+ "▁Kn ow",
+ "▁t rib",
+ "▁tr ib",
+ "▁tri b",
+ "▁r és",
+ "▁ré s",
+ "▁s table",
+ "▁st able",
+ "▁sta ble",
+ "▁stab le",
+ "▁ stable",
+ "▁W ould",
+ "▁Wo uld",
+ "▁ Would",
+ "▁break ing",
+ "▁bre aking",
+ "▁ breaking",
+ "▁me al",
+ "▁p hen",
+ "▁ph en",
+ "▁f el",
+ "▁fe l",
+ "▁ fel",
+ "▁F red",
+ "▁Fr ed",
+ "▁Fre d",
+ "Aut hor",
+ "Auth or",
+ "▁c apture",
+ "▁capt ure",
+ "▁ capture",
+ "op ts",
+ "opt s",
+ "o pts",
+ "▁every where",
+ "▁s que",
+ "▁squ e",
+ "▁sq ue",
+ "▁m oder",
+ "▁mod er",
+ "▁mo der",
+ "▁mode r",
+ "set up",
+ "▁S upp",
+ "▁Su pp",
+ "▁Sup p",
+ "▁ Supp",
+ "▁when ever",
+ "▁whe never",
+ "{ (",
+ "wa rt",
+ "war t",
+ "w art",
+ "▁t oe",
+ "▁to e",
+ "Pre fix",
+ "Pref ix",
+ "P refix",
+ "ho u",
+ "h ou",
+ "ga ge",
+ "g age",
+ "> \"",
+ "▁f rag",
+ "▁fr ag",
+ "▁fra g",
+ "▁ frag",
+ "▁The orem",
+ "mem ory",
+ "▁cont ents",
+ "▁content s",
+ "▁conten ts",
+ "▁ contents",
+ "do cs",
+ "doc s",
+ "} '",
+ "▁Ir ish",
+ "The n",
+ "Th en",
+ "T hen",
+ "aa ts",
+ "aat s",
+ "a ats",
+ "Sa ve",
+ "S ave",
+ "▁a gency",
+ "▁ag ency",
+ "▁и ме",
+ "▁им е",
+ "до ва",
+ "дов а",
+ "▁F unction",
+ "▁Fun ction",
+ "▁ Function",
+ "N N",
+ "dest roy",
+ "▁M essage",
+ "▁Mess age",
+ "▁ Message",
+ "▁c ancel",
+ "▁can cel",
+ "▁ cancel",
+ "▁super ior",
+ "▁e c",
+ "▁ ec",
+ "▁liter ature",
+ "▁P ART",
+ "▁PA RT",
+ "▁PAR T",
+ "▁ PART",
+ "I l",
+ "▁C ab",
+ "▁Ca b",
+ "eng ine",
+ "▁b asket",
+ "▁bas ket",
+ "wor th",
+ "wort h",
+ "w orth",
+ "▁S el",
+ "▁Se l",
+ "f etch",
+ "▁St adt",
+ "▁Stad t",
+ "▁Sta dt",
+ "▁К и",
+ "▁con j",
+ "▁se iner",
+ "▁sein er",
+ "▁seine r",
+ "▁sei ner",
+ "▁conf irmed",
+ "▁confirm ed",
+ "▁Ar gent",
+ "▁Arg ent",
+ "am ar",
+ "ama r",
+ "a mar",
+ "pgf path",
+ "▁strugg le",
+ "Pat tern",
+ "▁M iddle",
+ "it an",
+ "ita n",
+ "i tan",
+ "▁m oon",
+ "▁mo on",
+ "or ough",
+ "oro ugh",
+ "o rough",
+ "▁Cath olic",
+ "▁str uck",
+ "▁stru ck",
+ "] ->",
+ "▁we apon",
+ "▁weap on",
+ "▁su bst",
+ "▁sub st",
+ "▁subs t",
+ "▁inst ructions",
+ "▁instruct ions",
+ "▁instruction s",
+ "▁occ as",
+ "▁oc cas",
+ "prote cted",
+ "▁L ess",
+ "▁Le ss",
+ "▁Les s",
+ "▁ Less",
+ "▁b atch",
+ "▁bat ch",
+ "▁ batch",
+ "▁con tra",
+ "▁cont ra",
+ "▁contr a",
+ "▁de ck",
+ "▁dec k",
+ "▁ deck",
+ "▁ign ored",
+ "▁ignore d",
+ "▁ignor ed",
+ "▁ref used",
+ "▁refuse d",
+ "tr igger",
+ "▁crim inal",
+ "G A",
+ "ol ly",
+ "oll y",
+ "▁B ell",
+ "▁Be ll",
+ "▁Bel l",
+ "▁ Ю",
+ "for ward",
+ "▁p refix",
+ "▁pre fix",
+ "▁pref ix",
+ "▁ prefix",
+ "▁im mediate",
+ "▁immedi ate",
+ "▁as signed",
+ "▁ass igned",
+ "▁assign ed",
+ "▁e lected",
+ "▁elect ed",
+ "▁ele cted",
+ "▁to night",
+ "▁ton ight",
+ "▁D ies",
+ "▁Die s",
+ "▁Di es",
+ "▁B each",
+ "▁Be ach",
+ "▁pre ced",
+ "▁prec ed",
+ "ow ał",
+ "owa ł",
+ "▁gal ax",
+ "▁log ic",
+ "en za",
+ "enz a",
+ "▁Cap tain",
+ "▁Capt ain",
+ "▁H ay",
+ "▁Ha y",
+ "▁f acts",
+ "▁fact s",
+ "▁fac ts",
+ "▁н и",
+ "▁ ни",
+ "t é",
+ "▁s b",
+ "▁ sb",
+ "op ed",
+ "ope d",
+ "o ped",
+ "▁com bat",
+ "▁comb at",
+ "▁expl ore",
+ "▁explo re",
+ "▁( -",
+ "▁ (-",
+ "Load er",
+ "Lo ader",
+ "▁Wil son",
+ "▁l ocked",
+ "▁loc ked",
+ "▁lock ed",
+ "▁ locked",
+ ": ",
+ "▁O d",
+ "▁P rote",
+ "▁Pro te",
+ "▁Pr ote",
+ "▁ Prote",
+ "▁dis abled",
+ "▁disable d",
+ "▁ disabled",
+ "▁h atte",
+ "▁hat te",
+ "▁sh out",
+ "▁con structor",
+ "▁construct or",
+ "▁constru ctor",
+ "▁ constructor",
+ "б і",
+ "▁t ras",
+ "▁tr as",
+ "▁tra s",
+ "▁ tras",
+ "▁F ather",
+ "▁Fa ther",
+ "▁Fat her",
+ "▁ad j",
+ "▁ adj",
+ "▁Carol ina",
+ "▁F ood",
+ "▁Fo od",
+ "ba d",
+ "b ad",
+ "at ore",
+ "ator e",
+ "ato re",
+ "param eters",
+ "parameter s",
+ "▁F ull",
+ "▁Fu ll",
+ "▁ Full",
+ "[ -",
+ "▁\" #",
+ "▁T ry",
+ "▁Tr y",
+ "▁ Try",
+ "сь кої",
+ "сько ї",
+ "▁ex haust",
+ "▁sc roll",
+ "▁scr oll",
+ "▁ scroll",
+ "_ ;",
+ "Wh o",
+ "W ho",
+ "▁deliver ed",
+ "▁re ferred",
+ "▁refer red",
+ "▁pro spect",
+ "▁pros pect",
+ "sc an",
+ "s can",
+ "▁mod ified",
+ "▁ modified",
+ "Gener ator",
+ "▁ex cess",
+ "▁exc ess",
+ "▁k g",
+ "▁ kg",
+ "ze t",
+ "z et",
+ "ic z",
+ "i cz",
+ "clip se",
+ "cli pse",
+ "▁t ank",
+ "▁tan k",
+ "▁g uns",
+ "▁gu ns",
+ "▁gun s",
+ "▁G es",
+ "▁Ge s",
+ "in ton",
+ "int on",
+ "into n",
+ "▁Wed nesday",
+ "▁main ly",
+ "par ser",
+ "parse r",
+ "pars er",
+ "▁effect ively",
+ "▁effective ly",
+ "▁К у",
+ "▁res ident",
+ "▁resid ent",
+ "▁L i",
+ "▁ Li",
+ "▁f lying",
+ "▁fl ying",
+ "▁fly ing",
+ "▁may or",
+ "▁mayo r",
+ "ü h",
+ "ut a",
+ "u ta",
+ "▁col our",
+ "▁air craft",
+ "ter ior",
+ "te rior",
+ "n r",
+ "▁ke eps",
+ "▁keep s",
+ "fa n",
+ "f an",
+ "▁sh irt",
+ "▁ shirt",
+ "Com par",
+ "Comp ar",
+ "▁E th",
+ "▁Et h",
+ "Ma c",
+ "M ac",
+ "cle an",
+ "c lean",
+ "sl ice",
+ "cz y",
+ "c zy",
+ "▁g ender",
+ "▁gen der",
+ "▁ge nder",
+ "▁ gender",
+ "▁b utter",
+ "▁but ter",
+ "▁butt er",
+ "AU T",
+ "A UT",
+ "▁E lement",
+ "▁El ement",
+ "▁Ele ment",
+ "▁ Element",
+ "Fi n",
+ "F in",
+ "dm a",
+ "d ma",
+ "sam ple",
+ "s ample",
+ "Reg istry",
+ "▁class ic",
+ "▁dr ove",
+ "▁dro ve",
+ "p b",
+ "def ined",
+ "define d",
+ "d efined",
+ "▁re ward",
+ "▁r eward",
+ "ya l",
+ "y al",
+ "]) ,",
+ "] ),",
+ "▁B AS",
+ "▁BA S",
+ "▁hy per",
+ "▁hyp er",
+ "▁ hyper",
+ "▁Н и",
+ "▁) .",
+ "▁ ).",
+ "Ps i",
+ "P si",
+ "▁ent ries",
+ "▁entr ies",
+ "▁ entries",
+ "▁King dom",
+ "▁S ong",
+ "▁So ng",
+ "▁Son g",
+ "▁prom pt",
+ "cent ering",
+ "center ing",
+ "▁H olly",
+ "▁Hol ly",
+ "▁Holl y",
+ "em an",
+ "ema n",
+ "e man",
+ "▁pain ting",
+ "▁paint ing",
+ "▁form ation",
+ "▁format ion",
+ "▁ formation",
+ "▁Re quest",
+ "▁Requ est",
+ "▁ Request",
+ "cont roller",
+ "control ler",
+ "Reg ion",
+ "P Y",
+ "id ades",
+ "ida des",
+ "idad es",
+ "idade s",
+ "T L",
+ "▁dis able",
+ "▁ disable",
+ "▁re in",
+ "ri cal",
+ "ric al",
+ "r ical",
+ "\" \r",
+ "% )",
+ "▁S ab",
+ "▁Sa b",
+ "▁With out",
+ "▁ Without",
+ "Se rv",
+ "Ser v",
+ "S erv",
+ "▁Sh ort",
+ "▁ Short",
+ "▁ ю",
+ "▁re sc",
+ "▁r esc",
+ "▁res c",
+ "▁ resc",
+ "▁pattern s",
+ "▁Array List",
+ "▁ ArrayList",
+ "sym bol",
+ "s ymbol",
+ "ac o",
+ "a co",
+ "▁H om",
+ "▁Ho m",
+ "▁ Hom",
+ "he lp",
+ "hel p",
+ "▁h asta",
+ "▁has ta",
+ "▁ha sta",
+ "▁hast a",
+ "▁inst alled",
+ "▁install ed",
+ "at ie",
+ "ati e",
+ "▁vis ited",
+ "▁visit ed",
+ "▁Б е",
+ "){ \\",
+ ") {\\",
+ "▁des de",
+ "J ECT",
+ "▁d rew",
+ "▁dr ew",
+ "▁dre w",
+ "▁St ock",
+ "▁Sto ck",
+ "▁C ru",
+ "▁Cr u",
+ "DE F",
+ "D EF",
+ "ob by",
+ "obb y",
+ "iz able",
+ "iza ble",
+ "og ether",
+ "oge ther",
+ "▁a ber",
+ "▁ab er",
+ "▁d an",
+ "▁da n",
+ "▁ dan",
+ "al is",
+ "ali s",
+ "ta il",
+ "t ail",
+ "▁ex pressed",
+ "▁exp ressed",
+ "▁express ed",
+ "▁expr essed",
+ "▁A ccess",
+ "▁Acc ess",
+ "▁Ac cess",
+ "▁ Access",
+ "Se g",
+ "S eg",
+ "▁L ib",
+ "▁Li b",
+ "▁ Lib",
+ "▁sup ports",
+ "▁support s",
+ "▁supp orts",
+ "back ground",
+ "▁comm une",
+ "▁commun e",
+ "cal led",
+ "call ed",
+ "c alled",
+ "▁print f",
+ "▁prin tf",
+ "▁ printf",
+ "▁Pr ince",
+ "▁Prin ce",
+ "ни те",
+ "de pend",
+ "dep end",
+ "▁d els",
+ "▁de ls",
+ "▁del s",
+ "ne ur",
+ "n eur",
+ "▁recomm ended",
+ "▁recommend ed",
+ "▁found ed",
+ "▁mark ets",
+ "▁market s",
+ "▁destroy ed",
+ "▁ab stract",
+ "▁abs tract",
+ "▁ abstract",
+ "▁s erie",
+ "▁se rie",
+ "▁ser ie",
+ "▁ serie",
+ "▁D un",
+ "▁Du n",
+ "Te rm",
+ "T erm",
+ "▁p ortion",
+ "▁port ion",
+ "ad apter",
+ "is set",
+ "iss et",
+ "isse t",
+ "че ски",
+ "▁in teger",
+ "▁inte ger",
+ "▁ integer",
+ "▁return ing",
+ "en ties",
+ "ent ies",
+ "enti es",
+ "▁F air",
+ "▁Fa ir",
+ "▁U SB",
+ "▁US B",
+ "▁ USB",
+ "▁P rice",
+ "▁Pr ice",
+ "▁Pri ce",
+ "▁ Price",
+ "ig ate",
+ "iga te",
+ "i gate",
+ "▁sett led",
+ "▁settle d",
+ "({ \\",
+ "( {\\",
+ "ne k",
+ "n ek",
+ "▁the rm",
+ "▁th erm",
+ "▁ther m",
+ "▁c ig",
+ "▁ci g",
+ "án y",
+ "á ny",
+ "▁invest igation",
+ "▁investig ation",
+ "om eter",
+ "ome ter",
+ "omet er",
+ "SU P",
+ "S UP",
+ "So me",
+ "Som e",
+ "S ome",
+ "si ng",
+ "sin g",
+ "s ing",
+ "Con stant",
+ "Const ant",
+ "▁re tail",
+ "▁ret ail",
+ "ż y",
+ "▁dr inking",
+ "▁drink ing",
+ "▁In vest",
+ "▁Inv est",
+ "S V",
+ "ig inal",
+ "igin al",
+ "igi nal",
+ "▁B ow",
+ "▁Bo w",
+ "{{ \\",
+ "{ {\\",
+ "▁ass istance",
+ "▁assist ance",
+ "▁intel lect",
+ "IN IT",
+ "au g",
+ "a ug",
+ "▁Le on",
+ "▁Leo n",
+ "Su r",
+ "S ur",
+ "▁ad mit",
+ "▁adm it",
+ "▁Com mand",
+ "▁Comm and",
+ "▁ Command",
+ "il les",
+ "ill es",
+ "ille s",
+ "ro v",
+ "r ov",
+ "▁o h",
+ "▁ oh",
+ "▁n ão",
+ "▁mat ching",
+ "▁match ing",
+ "▁g enu",
+ "▁gen u",
+ "▁ge nu",
+ "▁O x",
+ "т ся",
+ "not ation",
+ "G O",
+ "▁N ap",
+ "▁Na p",
+ "▁ver ify",
+ "▁ verify",
+ "▁aus si",
+ "▁auss i",
+ "Date Time",
+ "▁su itable",
+ "▁suit able",
+ "▁ind icate",
+ "▁indic ate",
+ "▁L ive",
+ "▁Li ve",
+ "▁Liv e",
+ "▁ Live",
+ "Fe ature",
+ "▁tr acks",
+ "▁track s",
+ "▁tra cks",
+ "▁has n",
+ "▁ha sn",
+ "▁J ava",
+ "▁Ja va",
+ "▁ Java",
+ "▁close ly",
+ "▁clos ely",
+ "▁D ad",
+ "▁Da d",
+ "ce ive",
+ "▁Mar ket",
+ "▁Mark et",
+ "ag y",
+ "a gy",
+ "▁\" -",
+ "aw n",
+ "a wn",
+ "st ell",
+ "ste ll",
+ "pt on",
+ "pto n",
+ "p ton",
+ "ze it",
+ "▁V ector",
+ "▁Ve ctor",
+ "▁Vec tor",
+ "▁ Vector",
+ "▁M AX",
+ "▁MA X",
+ "▁ MAX",
+ "▁F ederal",
+ "▁Feder al",
+ "▁Fed eral",
+ "wa ll",
+ "wal l",
+ "w all",
+ "▁J en",
+ "▁Je n",
+ "de lay",
+ "del ay",
+ "▁lim its",
+ "▁limit s",
+ "▁ limits",
+ "▁Q uest",
+ "▁Qu est",
+ "▁Que st",
+ "▁ Quest",
+ "C am",
+ "▁F el",
+ "▁Fe l",
+ "write r",
+ "wr iter",
+ "writ er",
+ "w riter",
+ "L P",
+ "▁m oves",
+ "▁mov es",
+ "▁move s",
+ "▁mo ves",
+ "▁Ex ecut",
+ "▁ Execut",
+ "▁D B",
+ "▁ DB",
+ "ok er",
+ "oke r",
+ "o ker",
+ "sc ribe",
+ "scri be",
+ "scr ibe",
+ "scrib e",
+ "el ijk",
+ "elij k",
+ "eli jk",
+ "Const ants",
+ "Constant s",
+ "Add r",
+ "Ad dr",
+ "▁} }",
+ "▁ }}",
+ "▁ch annels",
+ "▁channel s",
+ "▁ channels",
+ "i y",
+ "rior ity",
+ "▁tr ading",
+ "▁trad ing",
+ "▁tra ding",
+ "▁fac ilities",
+ "▁facil ities",
+ "▁P ack",
+ "▁Pa ck",
+ "▁Pac k",
+ "▁ Pack",
+ "▁s ys",
+ "▁sy s",
+ "▁ sys",
+ "▁m eta",
+ "▁me ta",
+ "▁met a",
+ "▁ meta",
+ "▁est imate",
+ "▁estim ate",
+ "▁L ater",
+ "▁La ter",
+ "▁Lat er",
+ "▁Late r",
+ "iss ue",
+ "▁H aving",
+ "▁Ha ving",
+ "▁Hav ing",
+ "▁g uest",
+ "▁gu est",
+ "▁no body",
+ "▁nob ody",
+ "dep th",
+ "▁z ostał",
+ "пе ра",
+ "пер а",
+ ")} \\",
+ ") }\\",
+ "b g",
+ "▁Tw itter",
+ "▁dark ness",
+ "j pg",
+ "con tr",
+ "cont r",
+ "ker nel",
+ "kern el",
+ "k ernel",
+ "] \\",
+ "▁ext end",
+ "▁ extend",
+ "ro c",
+ "r oc",
+ "NE T",
+ "N ET",
+ "MS G",
+ "M SG",
+ "▁b urst",
+ "▁bur st",
+ "▁re pair",
+ "▁rep air",
+ "▁f etch",
+ "▁fet ch",
+ "▁ fetch",
+ "ie g",
+ "i eg",
+ "ú s",
+ "Sc reen",
+ "S creen",
+ "ble m",
+ "bl em",
+ "b lem",
+ "App Compat",
+ "▁ch ap",
+ "▁cha p",
+ "▁ chap",
+ "EL D",
+ "E LD",
+ "▁P enn",
+ "▁Pe nn",
+ "▁Pen n",
+ "▁prom ote",
+ "▁promot e",
+ "▁U kr",
+ "ar est",
+ "are st",
+ "ares t",
+ "a rest",
+ "▁s amples",
+ "▁sam ples",
+ "▁sample s",
+ "▁ samples",
+ "▁G reek",
+ "▁Gre ek",
+ "▁Gree k",
+ "▁con stru",
+ "▁const ru",
+ "▁constr u",
+ "▁un iverse",
+ "▁univers e",
+ "elij ke",
+ "elijk e",
+ "▁pre ferred",
+ "▁prefer red",
+ "▁Д е",
+ "▁I ra",
+ "▁Ir a",
+ "▁d ow",
+ "▁do w",
+ "ag ues",
+ "ague s",
+ "agu es",
+ "HE RE",
+ "HER E",
+ "H ERE",
+ "▁exper ts",
+ "▁exp erts",
+ "▁expert s",
+ "Pro tocol",
+ "Proto col",
+ "PI O",
+ "P IO",
+ "▁n az",
+ "▁na z",
+ "▁K h",
+ "hö r",
+ "h ör",
+ "▁dist ingu",
+ "▁B Y",
+ "▁ BY",
+ "▁se ine",
+ "▁sein e",
+ "▁sei ne",
+ "ep ing",
+ "e ping",
+ "▁fair ly",
+ "▁Me an",
+ "ix er",
+ "in si",
+ "ins i",
+ "▁author s",
+ "▁auth ors",
+ "** .",
+ "* *.",
+ "A I",
+ "▁ed ges",
+ "▁edge s",
+ "▁ edges",
+ "▁shoot ing",
+ "Ad min",
+ "▁m aps",
+ "▁map s",
+ "▁ma ps",
+ "▁ maps",
+ "ch ant",
+ "chan t",
+ "cha nt",
+ "▁CO VID",
+ "▁link ed",
+ "▁lin ked",
+ "▁ linked",
+ "▁s ke",
+ "▁sk e",
+ "▁ ske",
+ "▁power s",
+ "▁pow ers",
+ "á d",
+ "▁stom ach",
+ "▁us age",
+ "▁ usage",
+ "▁def end",
+ "▁defe nd",
+ "▁s ustain",
+ "▁sus tain",
+ "▁sust ain",
+ "▁up dates",
+ "▁update s",
+ "▁as sign",
+ "▁ass ign",
+ "▁ assign",
+ "H L",
+ "▁S ea",
+ "▁Se a",
+ "▁dis cipl",
+ "V ideo",
+ "▁Ch ief",
+ "▁Chi ef",
+ "▁b unch",
+ "▁Ob ama",
+ "ni s",
+ "n is",
+ "vo r",
+ "v or",
+ "▁ag ents",
+ "▁agent s",
+ "ca s",
+ "c as",
+ "ch ter",
+ "cht er",
+ "chte r",
+ "▁gl anced",
+ "▁glance d",
+ "support ed",
+ "supp orted",
+ "▁Cons ider",
+ "▁Every one",
+ "▁l ect",
+ "▁le ct",
+ "▁ lect",
+ "▁St one",
+ "▁Sto ne",
+ "▁J am",
+ "▁Ja m",
+ "og ram",
+ "o gram",
+ "form ance",
+ "▁\\ \"",
+ "▁ \\\"",
+ "▁p atch",
+ "▁pat ch",
+ "▁ patch",
+ "▁v it",
+ "▁vi t",
+ "Po wer",
+ "P ower",
+ "▁hard er",
+ "▁har der",
+ "An al",
+ "A nal",
+ "▁des ired",
+ "▁desire d",
+ "▁j ug",
+ "▁ju g",
+ "▁support ing",
+ "D U",
+ "]] ,",
+ "] ],",
+ "▁Ad ministr",
+ "▁Admin istr",
+ "uck y",
+ "uc ky",
+ "▁cont roller",
+ "▁control ler",
+ "▁ controller",
+ "▁iss ued",
+ "▁issue d",
+ "▁S in",
+ "▁Si n",
+ "▁aff ili",
+ "▁part ners",
+ "▁partner s",
+ "cd ots",
+ "cdot s",
+ "c dots",
+ "ct ic",
+ "C ar",
+ "▁N Y",
+ "▁ NY",
+ "▁p riority",
+ "▁prior ity",
+ "▁ priority",
+ "or iginal",
+ "orig inal",
+ "origin al",
+ "S ql",
+ "▁decl ared",
+ "▁declare d",
+ "▁declar ed",
+ "▁Hot el",
+ "▁b rowser",
+ "▁brow ser",
+ "▁brows er",
+ "▁ browser",
+ "▁gr ande",
+ "▁grand e",
+ "▁gran de",
+ "▁gra nde",
+ "}^ \\",
+ "} ^\\",
+ "bo w",
+ "b ow",
+ "▁accom mod",
+ "Direct ory",
+ "▁suff ering",
+ "▁suffer ing",
+ "▁log ger",
+ "▁ logger",
+ "▁break fast",
+ "ul i",
+ "u li",
+ "▁b oot",
+ "▁bo ot",
+ "▁ boot",
+ "▁contribut ion",
+ "NE SS",
+ "▁T en",
+ "▁Te n",
+ "▁ Ten",
+ "sem ble",
+ "semb le",
+ "sembl e",
+ "▁h ousing",
+ "▁hous ing",
+ "▁ho using",
+ "R aw",
+ "AN CE",
+ "▁П ри",
+ "▁b rit",
+ "▁br it",
+ "▁ brit",
+ "es sa",
+ "ess a",
+ "in son",
+ "ins on",
+ "▁B all",
+ "▁Ba ll",
+ "▁Bal l",
+ "en tes",
+ "ent es",
+ "ente s",
+ "▁B ra",
+ "▁Br a",
+ "sc ore",
+ "s core",
+ "GE R",
+ "G ER",
+ "ro ute",
+ "rou te",
+ "r oute",
+ "ap sed",
+ "aps ed",
+ "apse d",
+ "ро й",
+ "di ff",
+ "d iff",
+ "▁broad cast",
+ "▁t ar",
+ "▁ta r",
+ "▁ tar",
+ "▁de light",
+ "▁del ight",
+ ") ?",
+ "ch ester",
+ "che ster",
+ "ches ter",
+ "Pl atform",
+ "▁emer gency",
+ "▁c es",
+ "▁ce s",
+ "▁ ces",
+ "ner ship",
+ "ners hip",
+ "n ership",
+ "▁sit uations",
+ "▁situ ations",
+ "▁situation s",
+ "▁famil jen",
+ "▁G eb",
+ "▁Ge b",
+ "en ta",
+ "ent a",
+ "ú blic",
+ "▁P lace",
+ "▁Pl ace",
+ "▁ Place",
+ "IL L",
+ "I LL",
+ "▁m arch",
+ "▁mar ch",
+ "▁fundament al",
+ "att ributes",
+ "attribute s",
+ "кт и",
+ "к ти",
+ "▁F u",
+ "F D",
+ "▁ра с",
+ "▁academ ic",
+ "pr es",
+ "pre s",
+ "p res",
+ "▁r ising",
+ "▁ri sing",
+ "▁ris ing",
+ "▁B raz",
+ "▁Br az",
+ "▁Bra z",
+ "▁rece iving",
+ "WAR N",
+ "▁jud g",
+ "▁necess arily",
+ "] =",
+ "▁deep ly",
+ "▁g ray",
+ "▁gr ay",
+ "▁gra y",
+ "▁ gray",
+ "He aders",
+ "Head ers",
+ "Header s",
+ "▁co al",
+ "\\ {",
+ "Mu t",
+ "M ut",
+ "ba ch",
+ "b ach",
+ "▁pro fit",
+ "▁prof it",
+ "▁ profit",
+ "во го",
+ "в ого",
+ "ig s",
+ "i gs",
+ "og rap",
+ "\"; \r",
+ "\" ;\r",
+ "▁adv oc",
+ "Gener ated",
+ "Generate d",
+ "ме ри",
+ "мер и",
+ "▁C ond",
+ "▁Con d",
+ "▁Co nd",
+ "▁ Cond",
+ "▁ag ric",
+ "BA SE",
+ "B ASE",
+ "▁arr ang",
+ "▁flow ers",
+ "▁flower s",
+ "i w",
+ "▁] ;",
+ "▁ ];",
+ "▁во й",
+ "▁ вой",
+ "ume rate",
+ "umer ate",
+ "▁i hr",
+ "▁ih r",
+ "▁п ар",
+ "▁па р",
+ "▁ пар",
+ "▁m ont",
+ "▁mon t",
+ "▁mo nt",
+ "▁ mont",
+ "wide hat",
+ "m g",
+ "▁b tn",
+ "▁bt n",
+ "▁ btn",
+ "▁b esk",
+ "▁be sk",
+ "▁bes k",
+ "▁act s",
+ "▁ac ts",
+ "▁ acts",
+ "ó s",
+ "~~ ~~",
+ "▁cur ve",
+ "▁curv e",
+ "l anguage",
+ "▁TR UE",
+ "▁ TRUE",
+ "▁cle aning",
+ "▁clean ing",
+ "Mat h",
+ "Ma th",
+ "M ath",
+ "▁reg ional",
+ "▁region al",
+ "▁est imated",
+ "▁estim ated",
+ "▁estimate d",
+ "ar ity",
+ "ari ty",
+ "ier ung",
+ "/ {",
+ "jan go",
+ "j ango",
+ "$ _",
+ "▁th rew",
+ "▁thr ew",
+ "r q",
+ "co p",
+ "c op",
+ "ner gy",
+ "▁Acc ount",
+ "▁Ac count",
+ "▁ Account",
+ "pa l",
+ "p al",
+ "▁N ic",
+ "▁Ni c",
+ "]) )",
+ "] ))",
+ "▁aw esome",
+ "▁L oad",
+ "▁Lo ad",
+ "▁ Load",
+ "un nel",
+ "unn el",
+ "▁r ows",
+ "▁ro ws",
+ "▁row s",
+ "▁ rows",
+ "▁for each",
+ "▁fore ach",
+ "▁fo reach",
+ "▁ foreach",
+ "▁P od",
+ "▁Po d",
+ "▁ Pod",
+ "▁E N",
+ "▁ EN",
+ "▁. =",
+ "ua te",
+ "u ate",
+ "frastr ucture",
+ "▁W atch",
+ "▁Wat ch",
+ "▁ Watch",
+ "St and",
+ "▁r outine",
+ "▁rout ine",
+ "▁p ic",
+ "▁pi c",
+ "▁ pic",
+ "hel per",
+ "help er",
+ "▁hor ses",
+ "▁horse s",
+ "▁hors es",
+ "▁requ ested",
+ "▁request ed",
+ "▁- --",
+ "▁-- -",
+ "▁ ---",
+ "bor der",
+ "b order",
+ "▁lif ted",
+ "▁lift ed",
+ "▁P ed",
+ "▁Pe d",
+ "Im port",
+ "Imp ort",
+ "љ е",
+ "▁Л и",
+ "▁m yst",
+ "▁my st",
+ "TH ER",
+ "THE R",
+ "T HER",
+ "▁A C",
+ "▁ AC",
+ "Pro xy",
+ "Pr oxy",
+ "pro v",
+ "pr ov",
+ "p rov",
+ "▁N ik",
+ "▁Ni k",
+ "he mat",
+ "hem at",
+ "h emat",
+ "он аль",
+ "она ль",
+ "о наль",
+ "▁\" .",
+ "▁ \".",
+ "ul ui",
+ "ulu i",
+ "▁impro ved",
+ "▁improve d",
+ "ie ren",
+ "ier en",
+ "iere n",
+ "i eren",
+ "oc olate",
+ "ocol ate",
+ "oco late",
+ "Sc he",
+ "Sch e",
+ "S che",
+ "un ic",
+ "uni c",
+ "u nic",
+ "▁Profess or",
+ "ie ler",
+ "iel er",
+ "iele r",
+ "i eler",
+ "▁d uration",
+ "▁dur ation",
+ "▁ duration",
+ "▁time out",
+ "▁ timeout",
+ "ho m",
+ "h om",
+ "▁l ux",
+ "▁lu x",
+ "▁t rab",
+ "▁tr ab",
+ "▁tra b",
+ "it ary",
+ "ita ry",
+ "itar y",
+ "њ е",
+ "▁insp ired",
+ "▁inspir ed",
+ "▁inspire d",
+ "}) \\",
+ "} )\\",
+ "is ely",
+ "ise ly",
+ "ial s",
+ "ia ls",
+ "i als",
+ "▁V or",
+ "▁Vo r",
+ "▁enh ance",
+ "▁l ucky",
+ "▁luck y",
+ "▁luc ky",
+ "W orld",
+ "el o",
+ "e lo",
+ "if iers",
+ "ifier s",
+ "ifi ers",
+ "▁f acing",
+ "▁fac ing",
+ "▁fa cing",
+ "▁appreci ate",
+ "▁ être",
+ "▁ben ch",
+ "▁ bench",
+ "at ted",
+ "att ed",
+ "atte d",
+ "gen ce",
+ "g ence",
+ "c ourse",
+ "▁t ub",
+ "▁tu b",
+ "▁l ors",
+ "▁lo rs",
+ "▁mis take",
+ "▁mist ake",
+ "no m",
+ "n om",
+ "▁p aus",
+ "▁pa us",
+ "▁\" \";",
+ "▁\"\" ;",
+ "▁su bs",
+ "▁sub s",
+ "▁st ato",
+ "▁stat o",
+ "▁sta to",
+ "$ )",
+ "▁g ay",
+ "▁ga y",
+ "or ry",
+ "orr y",
+ "▁veh icles",
+ "▁vehicle s",
+ "▁br ill",
+ "ma y",
+ "m ay",
+ "re sp",
+ "res p",
+ "r esp",
+ "▁w ore",
+ "▁wor e",
+ "▁wo re",
+ "j ą",
+ "b p",
+ "on el",
+ "one l",
+ "o nel",
+ "▁C R",
+ "▁ CR",
+ "▁di agn",
+ "▁dia gn",
+ "math sf",
+ "▁hol iday",
+ "▁achie ved",
+ "▁achieve d",
+ "▁{ '",
+ "▁ {'",
+ "▁Re source",
+ "▁Res ource",
+ "▁ Resource",
+ "▁h i",
+ "▁ hi",
+ "▁b ra",
+ "▁br a",
+ "▁ bra",
+ "▁CON DITION",
+ "ct r",
+ "c tr",
+ "▁W rite",
+ "▁Writ e",
+ "▁Wr ite",
+ "▁ Write",
+ "is hop",
+ "ish op",
+ "i shop",
+ "OL D",
+ "O LD",
+ "▁c pu",
+ "▁cp u",
+ "▁ cpu",
+ "▁occ urs",
+ "▁occur s",
+ "▁oc curs",
+ "ó ł",
+ "str aint",
+ "stra int",
+ "▁nu clear",
+ "▁nuc lear",
+ "▁nucle ar",
+ "Ar ea",
+ "Are a",
+ "A rea",
+ "cl uster",
+ "▁surround ing",
+ "▁J uan",
+ "▁Ju an",
+ "▁pr ima",
+ "▁prim a",
+ "▁pri ma",
+ "▁South ern",
+ "▁Sou thern",
+ "it ty",
+ "itt y",
+ "i tty",
+ "▁As sembly",
+ "▁ Assembly",
+ "el em",
+ "ele m",
+ "e lem",
+ "ad i",
+ "a di",
+ "ér al",
+ "éra l",
+ "é ral",
+ "▁W at",
+ "▁Wa t",
+ "▁R adio",
+ "▁Rad io",
+ "▁ Radio",
+ "▁g egen",
+ "▁ge gen",
+ "▁T ony",
+ "▁To ny",
+ "▁Ton y",
+ "pr essed",
+ "press ed",
+ "pres sed",
+ "p ressed",
+ "▁An ne",
+ "▁Ann e",
+ "▁N S",
+ "▁ NS",
+ "▁P ak",
+ "▁Pa k",
+ "▁C ivil",
+ "▁Ci vil",
+ "▁th rown",
+ "▁throw n",
+ "▁thr own",
+ "▁thro wn",
+ "NO NE",
+ "NON E",
+ "N ONE",
+ "▁p ump",
+ "▁pu mp",
+ "▁s olve",
+ "▁sol ve",
+ "EN ABLE",
+ "▁Ph ys",
+ "▁ Phys",
+ "▁] ,",
+ "▁ ],",
+ "PO SE",
+ "POS E",
+ "kt et",
+ "kte t",
+ "▁F ab",
+ "▁Fa b",
+ "valid ate",
+ "Iter ator",
+ "cond ition",
+ "re du",
+ "red u",
+ "r edu",
+ "▁neg oti",
+ "an no",
+ "ann o",
+ "▁s ans",
+ "▁sa ns",
+ "▁san s",
+ "▁U l",
+ "CH AR",
+ "▁ed ition",
+ "▁edit ion",
+ "▁spect rum",
+ "or ie",
+ "ori e",
+ "o rie",
+ "▁execut ion",
+ "▁exec ution",
+ "P lease",
+ "▁B O",
+ "▁ BO",
+ "UR N",
+ "▁c ow",
+ "▁co w",
+ "▁ cow",
+ "ст ан",
+ "ста н",
+ "с тан",
+ "istribut ion",
+ "Do main",
+ "Dom ain",
+ "▁re aders",
+ "▁read ers",
+ "▁reader s",
+ "▁cons umer",
+ "▁consum er",
+ "▁consume r",
+ "▁st yles",
+ "▁style s",
+ "▁sty les",
+ "▁ styles",
+ "en code",
+ "enc ode",
+ "▁C y",
+ "Com mon",
+ "Comm on",
+ "▁P rop",
+ "▁Pro p",
+ "▁Pr op",
+ "▁ Prop",
+ "▁ex ecute",
+ "▁execut e",
+ "▁exec ute",
+ "▁ execute",
+ "▁e q",
+ "▁ eq",
+ "▁vis itors",
+ "▁visit ors",
+ "▁visitor s",
+ "▁A mb",
+ "▁Am b",
+ "ud ad",
+ "uda d",
+ "q quad",
+ "▁C ert",
+ "▁Ce rt",
+ "▁Cer t",
+ "▁ Cert",
+ "▁t rop",
+ "▁tr op",
+ "▁tro p",
+ "▁yes terday",
+ "ta in",
+ "t ain",
+ "L D",
+ "at ro",
+ "atr o",
+ "▁incre ases",
+ "▁increase s",
+ "▁W ars",
+ "▁War s",
+ "▁Wa rs",
+ "ne d",
+ "n ed",
+ "be fore",
+ "b efore",
+ "au pt",
+ "a upt",
+ "▁E RR",
+ "▁ER R",
+ "▁ ERR",
+ "▁F ord",
+ "▁For d",
+ "▁Fo rd",
+ "▁d alla",
+ "▁da lla",
+ "▁dal la",
+ "▁dall a",
+ "UL AR",
+ "▁st rike",
+ "▁str ike",
+ "▁stri ke",
+ "Ar r",
+ "A rr",
+ "▁re covery",
+ "▁rec overy",
+ "▁recover y",
+ "▁Res ponse",
+ "▁ Response",
+ "▁strateg ies",
+ "▁і н",
+ "▁ ін",
+ "▁re ar",
+ "▁r ear",
+ "▁adult s",
+ "▁Н е",
+ "window s",
+ "wind ows",
+ "de cl",
+ "dec l",
+ "ol en",
+ "ole n",
+ "o len",
+ "▁J ord",
+ "▁Jo rd",
+ "▁K al",
+ "▁Ka l",
+ "▁c ui",
+ "▁cu i",
+ "▁П ро",
+ "▁S ever",
+ "▁Se ver",
+ "▁Sev er",
+ "▁a le",
+ "▁al e",
+ "▁ ale",
+ "▁pe ut",
+ "▁peu t",
+ "St ats",
+ "Stat s",
+ "▁R oss",
+ "▁Ro ss",
+ "▁Ros s",
+ "ar ten",
+ "art en",
+ "arte n",
+ "sh all",
+ "shal l",
+ "sha ll",
+ "s hall",
+ "▁ent ertain",
+ "▁enter tain",
+ "▁entert ain",
+ "▁par king",
+ "▁park ing",
+ "но ви",
+ "нов и",
+ "er re",
+ "err e",
+ "▁fun ding",
+ "▁fund ing",
+ "▁C le",
+ "▁Cl e",
+ "▁O t",
+ "un st",
+ "uns t",
+ "assert Equals",
+ "assertEqual s",
+ "▁c ancell",
+ "▁can cell",
+ "▁cancel l",
+ "TA G",
+ "T AG",
+ "▁E arly",
+ "▁Earl y",
+ "▁feed back",
+ "▁p and",
+ "▁pan d",
+ "▁pa nd",
+ "y o",
+ "▁mir ror",
+ "▁ver b",
+ "▁ve rb",
+ "▁ verb",
+ "▁high light",
+ "er ialize",
+ "erial ize",
+ "▁g rade",
+ "▁gr ade",
+ "▁grad e",
+ "▁gra de",
+ "▁ grade",
+ "ла сь",
+ "▁Br ook",
+ "▁Bro ok",
+ "▁L I",
+ "▁ LI",
+ "▁im plies",
+ "▁impl ies",
+ "▁e norm",
+ "▁en orm",
+ "aj ą",
+ "a ją",
+ "▁W er",
+ "▁We r",
+ "aw ay",
+ "awa y",
+ "a way",
+ "▁machine s",
+ "▁mach ines",
+ "▁d ent",
+ "▁de nt",
+ "▁den t",
+ "Id x",
+ "I dx",
+ "▁t id",
+ "▁ti d",
+ "▁ tid",
+ ") \"",
+ "▁m ole",
+ "▁mo le",
+ "▁mol e",
+ "bo ld",
+ "bol d",
+ "b old",
+ "CO NT",
+ "CON T",
+ "C ONT",
+ "▁é p",
+ "▁ ép",
+ "▁cut ting",
+ "▁N eg",
+ "▁Ne g",
+ "▁ Neg",
+ "▁t ong",
+ "▁to ng",
+ "▁ton g",
+ "▁net works",
+ "▁network s",
+ "▁F all",
+ "▁Fa ll",
+ "▁Fal l",
+ "▁ Fall",
+ "gener ated",
+ "generate d",
+ "▁P ri",
+ "▁Pr i",
+ "UE ST",
+ "UES T",
+ "U EST",
+ "▁Be lg",
+ "▁Bel g",
+ "▁s heet",
+ "▁she et",
+ "▁ sheet",
+ "кс и",
+ "к си",
+ "▁ †",
+ "▁y eah",
+ "▁ye ah",
+ "▁Vict or",
+ "▁Vi ctor",
+ "▁Vic tor",
+ "▁R ub",
+ "▁Ru b",
+ "▁candid ates",
+ "▁candidate s",
+ "pr és",
+ "▁E U",
+ "et r",
+ "e tr",
+ "▁roll ed",
+ "▁ rolled",
+ "▁P as",
+ "▁Pa s",
+ "▁Ar thur",
+ "Ar ch",
+ "Arc h",
+ "▁M ann",
+ "▁Man n",
+ "▁Ma nn",
+ "Amer ican",
+ "America n",
+ "ze s",
+ "z es",
+ "in ners",
+ "inn ers",
+ "inner s",
+ "▁A uto",
+ "▁Aut o",
+ "▁Au to",
+ "▁ Auto",
+ "▁profess or",
+ "▁profes sor",
+ "▁) ;\r",
+ "▁); \r",
+ "▁ );\r",
+ "▁ad dr",
+ "▁add r",
+ "▁ addr",
+ "▁Med ical",
+ "▁Medic al",
+ "▁f ired",
+ "▁fire d",
+ "▁fi red",
+ "▁fir ed",
+ "▁C ore",
+ "▁Co re",
+ "▁Cor e",
+ "▁ Core",
+ "▁CON FIG",
+ "▁ CONFIG",
+ "▁s ql",
+ "▁sq l",
+ "▁ sql",
+ "▁Con serv",
+ "▁Cons erv",
+ "▁Conse rv",
+ "ic hen",
+ "ich en",
+ "iche n",
+ "i chen",
+ "Ver tex",
+ "Vert ex",
+ "▁H O",
+ "▁ HO",
+ "Y eah",
+ "No te",
+ "Not e",
+ "N ote",
+ "▁O K",
+ "▁ OK",
+ "mu s",
+ "m us",
+ "f ocus",
+ "aj a",
+ "a ja",
+ "r á",
+ "▁h ence",
+ "▁hen ce",
+ "▁execut ive",
+ "▁liqu id",
+ "uj e",
+ "u je",
+ "▁d riven",
+ "▁dr iven",
+ "▁dri ven",
+ "▁driv en",
+ "▁drive n",
+ "▁ driven",
+ "ig ue",
+ "igu e",
+ "i gue",
+ "▁W ik",
+ "▁Wi k",
+ "R ate",
+ "ra nd",
+ "ran d",
+ "r and",
+ "Result s",
+ "▁cop ies",
+ "▁t an",
+ "▁ta n",
+ "▁ tan",
+ "rit eria",
+ "rite ria",
+ "riter ia",
+ "en en",
+ "ene n",
+ "e nen",
+ "}_ \\",
+ "} _\\",
+ "▁po bl",
+ "▁pob l",
+ "▁sou thern",
+ "▁south ern",
+ "el n",
+ "e ln",
+ "▁z wei",
+ "▁zwe i",
+ "▁zw ei",
+ "▁con crete",
+ "▁CONDITION S",
+ "▁dream s",
+ "▁dre ams",
+ "▁min im",
+ "▁mi nim",
+ "▁mini m",
+ "▁em ployee",
+ "▁employ ee",
+ "▁n ap",
+ "▁na p",
+ "▁su spect",
+ "▁sus pect",
+ "▁susp ect",
+ "Mo use",
+ "M ouse",
+ "▁ther apy",
+ "▁therap y",
+ "av al",
+ "ava l",
+ "a val",
+ "▁An th",
+ "▁Ant h",
+ "ST ART",
+ "st ers",
+ "ster s",
+ "ste rs",
+ "s ters",
+ "ish ment",
+ "fin ite",
+ "W A",
+ "v y",
+ "▁m ood",
+ "▁mo od",
+ "com fort",
+ "▁s hr",
+ "▁sh r",
+ "▁dec ade",
+ "я бря",
+ "▁' #",
+ "▁d ot",
+ "▁do t",
+ "▁ dot",
+ "▁h ill",
+ "▁hi ll",
+ "▁ hill",
+ "ar ry",
+ "arr y",
+ "cat ch",
+ "c atch",
+ "▁j Query",
+ "▁ jQuery",
+ "▁corpor ate",
+ "▁BAS IS",
+ "▁appoint ed",
+ "▁em bar",
+ "▁emb ar",
+ "ograph ie",
+ "▁p ressed",
+ "▁pr essed",
+ "▁pres sed",
+ "▁press ed",
+ "▁ pressed",
+ "▁ch ampion",
+ "▁champ ion",
+ "em it",
+ "emi t",
+ "e mit",
+ "▁B ed",
+ "▁Be d",
+ "ва ння",
+ "ван ня",
+ "Gu i",
+ "G ui",
+ "▁P UR",
+ "▁ur ban",
+ "▁urb an",
+ "▁sent ence",
+ "bu ry",
+ "bur y",
+ "b ury",
+ "▁V ideo",
+ "▁ Video",
+ "▁regular ly",
+ "▁regul arly",
+ "v l",
+ "▁с лу",
+ "▁ слу",
+ "oc key",
+ "ock ey",
+ "ev in",
+ "e vin",
+ "ult ural",
+ "ultur al",
+ "▁pass age",
+ "▁со став",
+ "▁соста в",
+ "▁large ly",
+ "▁larg ely",
+ "or ters",
+ "ort ers",
+ "orter s",
+ "orte rs",
+ "▁conne ctions",
+ "▁connection s",
+ "▁connect ions",
+ "▁surpr ising",
+ "b c",
+ "▁strong ly",
+ "ans as",
+ "▁s ist",
+ "▁si st",
+ "▁ext reme",
+ "▁extrem e",
+ "▁extr eme",
+ "wh el",
+ "whe l",
+ "w hel",
+ "▁de aling",
+ "▁deal ing",
+ "ograph ic",
+ "▁Republic an",
+ "▁gr anted",
+ "▁gran ted",
+ "▁grant ed",
+ "▁C L",
+ "▁ CL",
+ "▁H ope",
+ "▁Ho pe",
+ "▁Hop e",
+ "less ly",
+ "▁u pload",
+ "▁up load",
+ "▁ upload",
+ "▁- \\",
+ "▁ -\\",
+ "ни ю",
+ "▁val uable",
+ "= [",
+ "Pr ice",
+ "P rice",
+ "iss ance",
+ "ie ns",
+ "ien s",
+ "i ens",
+ "he it",
+ "▁sugg ests",
+ "▁suggest s",
+ "с ло",
+ "▁j ur",
+ "▁ju r",
+ "} |",
+ "l p",
+ "▁inv ited",
+ "▁invite d",
+ "▁de riv",
+ "▁der iv",
+ "IM IT",
+ "I MIT",
+ "ra ss",
+ "ras s",
+ "r ass",
+ "▁in struct",
+ "▁inst ruct",
+ "▁instr uct",
+ "▁c ourses",
+ "▁cour ses",
+ "▁course s",
+ "▁cours es",
+ "ä ch",
+ "▁fif ty",
+ "▁fi fty",
+ "DE VICE",
+ "DEV ICE",
+ "AS H",
+ "A SH",
+ "▁h ip",
+ "▁hi p",
+ "▁ hip",
+ "Un known",
+ "▁C atalogue",
+ "▁Catal ogue",
+ "▁R oll",
+ "▁Ro ll",
+ "▁Rol l",
+ "▁ Roll",
+ "▁t ensor",
+ "▁ten sor",
+ "▁tens or",
+ "▁ tensor",
+ "be c",
+ "b ec",
+ "ét é",
+ "é té",
+ "Id entity",
+ "Ident ity",
+ "& \\",
+ "▁Step hen",
+ "▁Steph en",
+ "no des",
+ "node s",
+ "nod es",
+ "n odes",
+ "Di m",
+ "D im",
+ "▁cons ists",
+ "▁consist s",
+ "▁normal ly",
+ "▁norm ally",
+ "ub l",
+ "u bl",
+ "▁Pol ice",
+ "▁G ames",
+ "▁Game s",
+ "▁Ga mes",
+ "▁Gam es",
+ "fi ve",
+ "f ive",
+ "Ha ve",
+ "H ave",
+ "▁p adding",
+ "▁pad ding",
+ "▁ padding",
+ "er es",
+ "ere s",
+ "e res",
+ "an th",
+ "ant h",
+ "▁p uts",
+ "▁put s",
+ "▁pu ts",
+ "um inate",
+ "umin ate",
+ "umi nate",
+ "ov ie",
+ "ovi e",
+ "▁In dex",
+ "▁Ind ex",
+ "▁ Index",
+ "bl ue",
+ "Sc al",
+ "S cal",
+ "▁g iant",
+ "▁gi ant",
+ "T F",
+ "ps on",
+ "p son",
+ "▁vict im",
+ "▁vic tim",
+ "se rial",
+ "ser ial",
+ "s erial",
+ "▁S ym",
+ "▁Sy m",
+ "▁ Sym",
+ "Sing le",
+ "S ingle",
+ "▁m d",
+ "▁ md",
+ "▁att ended",
+ "▁attend ed",
+ "▁S tra",
+ "▁St ra",
+ "▁Str a",
+ "▁D ark",
+ "▁Dar k",
+ "▁ Dark",
+ ") |",
+ "▁s pan",
+ "▁sp an",
+ "▁ span",
+ "▁main tenance",
+ "▁b ind",
+ "▁bi nd",
+ "▁bin d",
+ "▁ bind",
+ "Be an",
+ "il arly",
+ "ilar ly",
+ "▁con vent",
+ "▁conv ent",
+ "▁conven t",
+ "▁conve nt",
+ "▁Jos é",
+ "ud d",
+ "u dd",
+ "▁p oly",
+ "▁pol y",
+ "▁po ly",
+ "▁ poly",
+ "▁i dx",
+ "▁id x",
+ "▁ idx",
+ "▁as ks",
+ "▁ask s",
+ "▁ent hus",
+ "▁s uck",
+ "▁su ck",
+ "▁suc k",
+ "▁C ou",
+ "▁Co u",
+ "▁Corpor ation",
+ "us ions",
+ "usion s",
+ "op her",
+ "oph er",
+ "o pher",
+ "▁sympt oms",
+ "▁Joh ann",
+ "▁п у",
+ "▁ пу",
+ "▁h tml",
+ "▁ html",
+ "▁p s",
+ "▁ ps",
+ "ear ing",
+ "ea ring",
+ "e aring",
+ "ge sch",
+ "ges ch",
+ "g esch",
+ "▁M other",
+ "▁Mo ther",
+ "▁Mot her",
+ "RE T",
+ "R ET",
+ "▁furn iture",
+ "P F",
+ "▁Gu ard",
+ "▁ Guard",
+ "pat tern",
+ "▁love ly",
+ "▁lov ely",
+ "al g",
+ "a lg",
+ "ed ly",
+ "se x",
+ "s ex",
+ "▁fin ds",
+ "▁find s",
+ "Bu f",
+ "B uf",
+ "▁на д",
+ "▁ над",
+ "▁к м",
+ "▁P or",
+ "▁Po r",
+ "С Р",
+ "En ter",
+ "Ent er",
+ "▁e sta",
+ "▁est a",
+ "▁es ta",
+ "▁ esta",
+ "▁т ре",
+ "▁ тре",
+ "▁\" *",
+ "▁F ox",
+ "▁Fo x",
+ "▁c ock",
+ "▁co ck",
+ "▁coc k",
+ "▁ cock",
+ "B undle",
+ "▁p uis",
+ "▁pu is",
+ "▁ puis",
+ "▁ann ounce",
+ "▁announ ce",
+ "▁g uid",
+ "▁gu id",
+ "▁ guid",
+ "check ed",
+ "ic ide",
+ "ici de",
+ "ne g",
+ "n eg",
+ "▁G il",
+ "▁Gi l",
+ "sc hen",
+ "sch en",
+ "sche n",
+ "s chen",
+ "olog ist",
+ "is o",
+ "i so",
+ "group s",
+ "gro ups",
+ "g roups",
+ "▁some body",
+ "Da y",
+ "D ay",
+ "tr as",
+ "tra s",
+ "t ras",
+ "▁comp act",
+ "▁organ ized",
+ "▁organiz ed",
+ "▁organize d",
+ "▁r oles",
+ "▁ro les",
+ "▁role s",
+ "▁h int",
+ "▁hi nt",
+ "▁ hint",
+ "▁s å",
+ "▁p ays",
+ "▁pay s",
+ "▁pa ys",
+ "▁С и",
+ "▁h oped",
+ "▁hope d",
+ "▁hop ed",
+ "▁ho ped",
+ "▁s ail",
+ "▁sa il",
+ "▁V ers",
+ "▁Ver s",
+ "▁Ve rs",
+ "▁ Vers",
+ "▁em br",
+ "▁emb r",
+ "▁b ot",
+ "▁bo t",
+ "▁ bot",
+ "▁ex ceed",
+ "▁exc eed",
+ "BA CK",
+ "B ACK",
+ "▁g aze",
+ "▁gaz e",
+ "▁ga ze",
+ "▁s pons",
+ "▁sp ons",
+ "▁spo ns",
+ "AS T",
+ "A ST",
+ "▁tor ch",
+ "▁ torch",
+ "▁news paper",
+ "▁newsp aper",
+ "▁D ist",
+ "▁Dis t",
+ "▁Di st",
+ "▁ Dist",
+ "▁b ass",
+ "▁bas s",
+ "▁ba ss",
+ "▁h anging",
+ "▁han ging",
+ "▁hang ing",
+ "▁e ars",
+ "▁ear s",
+ "▁ ears",
+ "ń sk",
+ "get Value",
+ "▁un us",
+ "▁E le",
+ "▁El e",
+ "serv ices",
+ "service s",
+ "s ervices",
+ "▁d ressed",
+ "▁dr essed",
+ "▁dress ed",
+ "la v",
+ "l av",
+ "▁п ла",
+ "▁ пла",
+ "Priv ate",
+ "P rivate",
+ "mi c",
+ "m ic",
+ "▁par ser",
+ "▁parse r",
+ "▁ parser",
+ "▁se ctions",
+ "▁section s",
+ "▁sect ions",
+ "▁ sections",
+ "▁f o",
+ "▁ fo",
+ "Err orf",
+ "Error f",
+ "in z",
+ "ör d",
+ "ö rd",
+ "▁m etric",
+ "▁met ric",
+ "▁ metric",
+ "UR I",
+ "U RI",
+ "▁v ice",
+ "▁vi ce",
+ "▁vic e",
+ "RE D",
+ "R ED",
+ "▁n ue",
+ "▁nu e",
+ "re vs",
+ "rev s",
+ "▁col lected",
+ "▁collect ed",
+ "▁colle cted",
+ "oo se",
+ "o ose",
+ "▁m ond",
+ "▁mon d",
+ "▁mo nd",
+ "▁ mond",
+ "▁n as",
+ "▁na s",
+ "▁ nas",
+ "▁На се",
+ "▁ å",
+ "Dr op",
+ "D rop",
+ "▁ab use",
+ "▁s ees",
+ "▁se es",
+ "▁see s",
+ "▁H ence",
+ "▁Hen ce",
+ "ex ec",
+ "}\\ ,",
+ "} \\,",
+ "▁ar bitr",
+ "▁Ap plication",
+ "▁ Application",
+ "f amily",
+ "ü d",
+ "▁mag netic",
+ "▁magn etic",
+ "▁magnet ic",
+ "▁new ly",
+ "▁re produ",
+ "▁rep rodu",
+ "▁writ ers",
+ "▁write rs",
+ "▁writer s",
+ "▁he aders",
+ "▁head ers",
+ "▁header s",
+ "▁ headers",
+ "š í",
+ "р т",
+ "YP E",
+ "Y PE",
+ "▁s chema",
+ "▁sch ema",
+ "▁sche ma",
+ "▁ schema",
+ "▁C e",
+ "▁Je ws",
+ "▁Jew s",
+ "▁Re cord",
+ "▁Rec ord",
+ "▁ Record",
+ "pre sent",
+ "pres ent",
+ "p resent",
+ "▁так же",
+ "▁label s",
+ "▁lab els",
+ "▁ labels",
+ "S ocket",
+ "▁equ ations",
+ "▁equation s",
+ "▁eq uations",
+ "▁medic ine",
+ "▁author ities",
+ "} `",
+ "ст ви",
+ "ств и",
+ "▁C orn",
+ "▁Co rn",
+ "▁Cor n",
+ "▁environment al",
+ "WAR E",
+ "WA RE",
+ "W ARE",
+ "Me r",
+ "M er",
+ "▁са мо",
+ "▁Techn ology",
+ "▁S af",
+ "▁Sa f",
+ "▁con n",
+ "▁co nn",
+ "▁ conn",
+ "▁U m",
+ "▁Pac ific",
+ "те л",
+ "ja n",
+ "j an",
+ "▁unc ertain",
+ "▁bel ief",
+ "▁belie f",
+ "co unter",
+ "count er",
+ "c ounter",
+ "to Be",
+ "IN S",
+ "I NS",
+ "we et",
+ "Li ght",
+ "L ight",
+ "pr imary",
+ "prim ary",
+ "▁feature d",
+ "▁feat ured",
+ "▁touch ed",
+ "▁tou ched",
+ "HT TP",
+ "▁t act",
+ "▁ta ct",
+ "pos itory",
+ "p ository",
+ "▁e ines",
+ "▁ein es",
+ "▁eine s",
+ "la ss",
+ "las s",
+ "l ass",
+ "сь ка",
+ "▁prz ez",
+ "▁prze z",
+ "▁f uer",
+ "▁fue r",
+ "▁fu er",
+ "▁exc iting",
+ "▁excit ing",
+ "▁C ub",
+ "▁Cu b",
+ "ag an",
+ "aga n",
+ "a gan",
+ "V O",
+ "▁' %",
+ "▁\\ {",
+ "▁ \\{",
+ "ub ble",
+ "▁F ol",
+ "▁Fo l",
+ "▁K ong",
+ "▁Kon g",
+ "▁Ko ng",
+ "▁ver sch",
+ "▁vers ch",
+ "FA IL",
+ "F AIL",
+ "▁na ar",
+ "ö s",
+ "sp eed",
+ "spe ed",
+ "s peed",
+ "▁terr itor",
+ "▁territo r",
+ "▁w rap",
+ "▁wr ap",
+ "▁ wrap",
+ "▁Jah re",
+ "▁Jahr e",
+ "▁Ja hre",
+ "le e",
+ "l ee",
+ "▁cross ed",
+ "res olve",
+ "▁s tim",
+ "▁st im",
+ "N ative",
+ "ur sor",
+ "urs or",
+ "Not Null",
+ "▁Al bert",
+ "▁Alber t",
+ "▁Alb ert",
+ "▁sign ature",
+ "▁ signature",
+ "▁R u",
+ "id as",
+ "ida s",
+ "i das",
+ "▁de cent",
+ "▁dec ent",
+ "▁dece nt",
+ "▁f aced",
+ "▁face d",
+ "▁fac ed",
+ "▁fa ced",
+ "▁ faced",
+ "▁ лю",
+ "▁Sp ain",
+ "▁res istance",
+ "▁resist ance",
+ "▁B rian",
+ "▁Br ian",
+ "kw args",
+ "▁inter val",
+ "▁ interval",
+ "▁Л е",
+ "▁ex plo",
+ "▁expl o",
+ "▁exp lo",
+ "▁s emi",
+ "▁se mi",
+ "▁sem i",
+ "▁wide ly",
+ "▁wid ely",
+ "d x",
+ "ko v",
+ "k ov",
+ "▁C ome",
+ "▁Com e",
+ "▁Co me",
+ "▁ Come",
+ "▁kn ife",
+ "As p",
+ "A sp",
+ "un o",
+ "u no",
+ "line to",
+ "lin eto",
+ "▁B und",
+ "▁Bu nd",
+ "▁Bun d",
+ "C ert",
+ "▁t odo",
+ "▁to do",
+ "▁tod o",
+ "ta gs",
+ "tag s",
+ "t ags",
+ "▁guarante e",
+ "▁v ital",
+ "▁vi tal",
+ "▁vit al",
+ "▁vita l",
+ "▁f ought",
+ "▁fou ght",
+ "▁E nv",
+ "▁En v",
+ "▁ Env",
+ "H D",
+ "Lo wer",
+ "Low er",
+ "L ower",
+ "T x",
+ "▁F a",
+ "▁ant icip",
+ "▁anti cip",
+ "Time r",
+ "Tim er",
+ "T imer",
+ "med iate",
+ "medi ate",
+ "media te",
+ "▁pro ven",
+ "▁pr oven",
+ "▁prov en",
+ "▁prove n",
+ "▁part ir",
+ "▁parti r",
+ "A E",
+ "cur sor",
+ "curs or",
+ "c ursor",
+ "▁wood en",
+ "▁wo oden",
+ "▁Cont act",
+ "▁ Contact",
+ "re gs",
+ "reg s",
+ "▁prov inc",
+ "▁provin c",
+ "▁D C",
+ "▁ DC",
+ "▁mem ories",
+ "▁memor ies",
+ "▁memo ries",
+ "▁f t",
+ "▁ ft",
+ "▁b attery",
+ "▁batter y",
+ "▁batt ery",
+ "▁bat tery",
+ "ute nant",
+ "uten ant",
+ "u tenant",
+ "Log in",
+ "Lo gin",
+ "ount ry",
+ "oun try",
+ "▁comp ens",
+ "operator name",
+ "▁Jac ob",
+ "ze d",
+ "z ed",
+ "AD DR",
+ "ADD R",
+ "▁qu ad",
+ "▁ quad",
+ "*) .",
+ "* ).",
+ "▁co at",
+ "▁f ir",
+ "▁fi r",
+ "▁Mich el",
+ "▁Mic hel",
+ "▁Mi chel",
+ "▁Miche l",
+ "▁Stand ard",
+ "▁ Standard",
+ "r f",
+ "me l",
+ "m el",
+ "▁co eff",
+ "▁Ira q",
+ "▁G iven",
+ "▁Gi ven",
+ "▁Give n",
+ "ни ма",
+ "ним а",
+ "▁F IT",
+ "▁FI T",
+ "▁p eu",
+ "▁pe u",
+ "▁i g",
+ "▁ ig",
+ "▁C ase",
+ "▁Cas e",
+ "▁Ca se",
+ "▁ Case",
+ "m é",
+ "▁par allel",
+ "▁ parallel",
+ "ci o",
+ "c io",
+ "ko w",
+ "k ow",
+ "▁institut ions",
+ "▁institution s",
+ "í cul",
+ "ab an",
+ "aba n",
+ "a ban",
+ "U X",
+ "▁Sa rah",
+ "▁Sar ah",
+ "▁Sara h",
+ "▁m és",
+ "▁mé s",
+ "▁at mos",
+ "▁atm os",
+ "▁slä ktet",
+ "▁br others",
+ "▁bro thers",
+ "▁brother s",
+ "▁want ing",
+ "aa aa",
+ "▁f est",
+ "▁fe st",
+ "= -",
+ "▁for ty",
+ "▁fort y",
+ "▁cre ates",
+ "▁create s",
+ "▁creat es",
+ "h h",
+ "▁And roid",
+ "▁Andr oid",
+ "▁ Android",
+ "an ches",
+ "anc hes",
+ "anch es",
+ "anche s",
+ "B T",
+ "up load",
+ "u pload",
+ "xi s",
+ "x is",
+ "H z",
+ "бо р",
+ "б ор",
+ "RA Y",
+ "R AY",
+ "nt il",
+ "n til",
+ "▁le aned",
+ "▁lean ed",
+ "un da",
+ "und a",
+ "▁ult imately",
+ "▁ultimate ly",
+ "▁t ok",
+ "▁to k",
+ "▁ tok",
+ "ne h",
+ "n eh",
+ "▁law yer",
+ "he nd",
+ "hen d",
+ "h end",
+ "▁V in",
+ "▁Vi n",
+ "▁fac ility",
+ "▁facil ity",
+ "▁l ikes",
+ "▁li kes",
+ "▁like s",
+ "▁lik es",
+ "en to",
+ "ent o",
+ "Node s",
+ "No des",
+ "N odes",
+ "▁entr ance",
+ "at to",
+ "att o",
+ "a tto",
+ "re tt",
+ "ret t",
+ "r ett",
+ "ac cept",
+ "th eme",
+ "the me",
+ "та н",
+ "т ан",
+ "os i",
+ "o si",
+ "▁{ },",
+ "▁{} ,",
+ "▁ {},",
+ "pgfpath lineto",
+ "go od",
+ "g ood",
+ "sl ot",
+ "s lot",
+ "▁in noc",
+ "▁inn oc",
+ "▁pro port",
+ "▁pr oport",
+ "▁prop ort",
+ "▁ar rive",
+ "▁arriv e",
+ "▁arr ive",
+ "é ho",
+ "▁p airs",
+ "▁pa irs",
+ "▁pair s",
+ "▁wr apped",
+ "▁wrap ped",
+ "▁un w",
+ "▁expl os",
+ "▁exp los",
+ "▁explo s",
+ "▁g el",
+ "▁ge l",
+ "▁ gel",
+ "W ill",
+ "▁Ze aland",
+ "ía s",
+ "í as",
+ "▁J r",
+ "▁F ra",
+ "▁Fr a",
+ "▁le git",
+ "▁leg it",
+ "▁il legal",
+ "к лю",
+ "▁t ort",
+ "▁to rt",
+ "▁tor t",
+ "▁p ron",
+ "▁pro n",
+ "▁pr on",
+ "F i",
+ "▁f org",
+ "▁for g",
+ "▁fo rg",
+ "ex port",
+ "exp ort",
+ "▁Child ren",
+ "▁ Children",
+ "▁A bs",
+ "▁Ab s",
+ "▁ Abs",
+ "▁S end",
+ "▁Se nd",
+ "▁Sen d",
+ "▁ Send",
+ "▁dis count",
+ "▁disc ount",
+ "▁disco unt",
+ "▁p oster",
+ "▁pos ter",
+ "▁po ster",
+ "▁post er",
+ "en ted",
+ "ent ed",
+ "ente d",
+ "an im",
+ "ani m",
+ "a nim",
+ "ve rb",
+ "ver b",
+ "st o",
+ "s to",
+ "▁B ible",
+ "▁Bi ble",
+ "pend ing",
+ "pen ding",
+ "p ending",
+ "▁P hot",
+ "▁Ph ot",
+ "st rap",
+ "str ap",
+ "stra p",
+ "ie ron",
+ "ier on",
+ "iero n",
+ "i eron",
+ "P G",
+ "cul ar",
+ "cu lar",
+ "c ular",
+ "cri t",
+ "cr it",
+ "c rit",
+ "ur d",
+ "u rd",
+ "EN O",
+ "E NO",
+ "▁nor thern",
+ "▁north ern",
+ "▁natural ly",
+ "▁natur ally",
+ "< '",
+ "we g",
+ "w eg",
+ "▁dr unk",
+ "▁D al",
+ "▁Da l",
+ "▁m ouse",
+ "▁mo use",
+ "▁mou se",
+ "▁ mouse",
+ "▁contin uous",
+ "▁continu ous",
+ "▁init ially",
+ "▁initial ly",
+ "▁initi ally",
+ "ag u",
+ "a gu",
+ "м пи",
+ "AN T",
+ "A NT",
+ "Di v",
+ "D iv",
+ "▁rec ording",
+ "▁record ing",
+ "Bin d",
+ "Bi nd",
+ "B ind",
+ "▁correct ly",
+ "init ial",
+ "▁R ights",
+ "▁Right s",
+ "▁deb ate",
+ "WR ITE",
+ "bu ilt",
+ "▁per mit",
+ "▁perm it",
+ "▁professional s",
+ "▁profession als",
+ "c v",
+ "▁D I",
+ "▁ DI",
+ "▁h anded",
+ "▁hand ed",
+ "▁han ded",
+ "▁ handed",
+ "▁C u",
+ "▁H ospital",
+ "▁besk revs",
+ "не й",
+ "н ей",
+ "но ст",
+ "▁anx iety",
+ "▁heav ily",
+ "▁V ar",
+ "▁Va r",
+ "▁ Var",
+ "▁dis pos",
+ "▁disp os",
+ "+ \"",
+ "▁E ver",
+ "▁Ev er",
+ "▁Eve r",
+ "iz on",
+ "izo n",
+ "i zon",
+ "▁oper ators",
+ "▁operator s",
+ "ne go",
+ "neg o",
+ "n ego",
+ "▁B ry",
+ "▁Br y",
+ "▁v otes",
+ "▁vo tes",
+ "▁vote s",
+ "▁vot es",
+ "iz ione",
+ "izi one",
+ "izio ne",
+ "i zione",
+ "▁ра й",
+ "▁fe at",
+ "▁ feat",
+ "▁w estern",
+ "▁west ern",
+ "▁ western",
+ "▁con front",
+ "▁strong er",
+ "▁ф а",
+ "▁ фа",
+ "st re",
+ "str e",
+ "s tre",
+ "▁Val id",
+ "▁ Valid",
+ "▁n ad",
+ "▁na d",
+ "▁check ing",
+ "▁bird s",
+ "▁North ern",
+ "▁Nor thern",
+ "▁int ention",
+ "▁intent ion",
+ "uc e",
+ "u ce",
+ "▁co vers",
+ "▁cover s",
+ "▁cov ers",
+ "▁wonder ing",
+ "▁Option al",
+ "▁Opt ional",
+ "▁ Optional",
+ "pro tocol",
+ "proto col",
+ "prot ocol",
+ "▁ag gress",
+ "— —",
+ "V ec",
+ "▁d ates",
+ "▁da tes",
+ "▁dat es",
+ "▁date s",
+ "▁ dates",
+ "qu ot",
+ "▁b om",
+ "▁bo m",
+ "▁s can",
+ "▁sc an",
+ "▁ scan",
+ "▁I tem",
+ "▁It em",
+ "▁ Item",
+ "▁N avy",
+ "▁Na vy",
+ "▁Nav y",
+ "▁G ran",
+ "▁Gr an",
+ "▁Gra n",
+ "▁every body",
+ "▁un expected",
+ "▁une xpected",
+ "▁di vor",
+ "▁div or",
+ "▁e ase",
+ "▁eas e",
+ "um bled",
+ "umb led",
+ "umble d",
+ "^ +",
+ "cu ss",
+ "c uss",
+ "▁p ale",
+ "▁pal e",
+ "▁pa le",
+ "▁In ga",
+ "▁Ing a",
+ "▁B road",
+ "▁Br oad",
+ "▁Bro ad",
+ "▁ Broad",
+ "▁Med ic",
+ "▁R oy",
+ "▁Ro y",
+ "▁I nn",
+ "▁In n",
+ "▁p ens",
+ "▁pe ns",
+ "▁pen s",
+ "P N",
+ ". :",
+ "▁princip le",
+ "▁let ting",
+ "▁lett ing",
+ "▁condu cted",
+ "▁conduct ed",
+ "F ALSE",
+ "▁O S",
+ "▁ OS",
+ "F ocus",
+ "▁measure d",
+ "▁meas ured",
+ "▁Dem ocratic",
+ "▁Democr atic",
+ "▁Democrat ic",
+ "Hi gh",
+ "H igh",
+ "▁p ré",
+ "▁pr é",
+ "en nes",
+ "enn es",
+ "enne s",
+ "▁ind icates",
+ "▁indic ates",
+ "▁indicate s",
+ "▁en ding",
+ "▁end ing",
+ "▁ ending",
+ "▁Sm all",
+ "▁ Small",
+ "▁< !--",
+ "▁