Update to v0.2 checkpoint

README.md

@@ -1,12 +1,21 @@
 ---
-library_name: transformers
 license: apache-2.0
+base_model: HuggingFaceTB/SmolLM-1.7B
+tags:
+- alignment-handbook
+- trl
+- sft
+datasets:
+- Magpie-Align/Magpie-Pro-300K-Filtered
+- bigcode/self-oss-instruct-sc2-exec-filter-50k
+- teknium/OpenHermes-2.5
+- HuggingFaceTB/everyday-conversations-llama3.1-2k
+library_name: transformers
 language:
 - en
 ---
 
-
-# SmolLM-Instruct
+# SmolLM-1.7B-Instruct
 
 <center>
     <img src="https://huggingface.co/datasets/HuggingFaceTB/images/resolve/main/banner_smol.png" alt="SmolLM" width="1100" height="600">
@@ -15,14 +24,35 @@ language:
 
 ## Model Summary
 
-SmolLM is a series of state-of-the-art small language models available in three sizes: 135M, 360M, and 1.7B parameters. These models are built on Cosmo-Corpus, a meticulously curated high-quality training dataset. Cosmo-Corpus includes Cosmopedia v2 (28B tokens of synthetic textbooks and stories generated by Mixtral), Python-Edu (4B tokens of educational Python samples from The Stack), and FineWeb-Edu (220B tokens of deduplicated educational web samples from FineWeb). For further details, we refer to our [blogpost](https://huggingface.co/blog/smollm).
+SmolLM is a series of small language models available in three sizes: 135M, 360M, and 1.7B parameters. 
+
+These models are pre-trained on [SmolLM-Corpus](https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus), a curated collection of high-quality educational and synthetic data designed for training LLMs. For further details, we refer to our [blogpost](https://huggingface.co/blog/smollm).
+
+To build SmolLM-Instruct, we finetuned the base models on publicly available datasets.
+
+## Changelog
+
+
+|Release|Description|
+|-|-|
+|v0.1| Initial release of SmolLM-Instruct. We finetune on the permissive subset of the [WebInstructSub](https://huggingface.co/datasets/TIGER-Lab/WebInstructSub) dataset, combined with [StarCoder2-Self-OSS-Instruct](https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k). Then, we perform DPO (Direct Preference Optimization) for one epoch on [HelpSteer](https://huggingface.co/datasets/nvidia/HelpSteer) for the 135M and 1.7B models, and [argilla/dpo-mix-7k](https://huggingface.co/datasets/argilla/dpo-mix-7k) for the 360M model.|
+|v0.2| We changed the finetuning mix to datasets more suitable for smol models. We train on a new dataset of 2k simple everyday conversations we generated by llama3.1-70B [everyday-conversations-llama3.1-2k](https://huggingface.co/datasets/HuggingFaceTB/everyday-conversations-llama3.1-2k/), [Magpie-Pro-300K-Filtered](https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-300K-Filtered), [StarCoder2-Self-OSS-Instruct](https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k), and a small subset of [OpenHermes-2.5](https://huggingface.co/datasets/teknium/OpenHermes-2.5)|
+
+v0.2 models are better at staying on topic and responding appropriately to standard prompts, such as greetings and questions about their role as AI assistants. SmolLM-360M-Instruct (v0.2) has a 63.3% win rate over SmolLM-360M-Instruct (v0.1) on AlpacaEval. You can find the details [here](https://huggingface.co/datasets/HuggingFaceTB/alpaca_eval_details/).
 
-To build SmolLM-Instruct, we instruction tuned the models using publicly available permissive instruction datasets. We trained all three models for one epoch on the permissive subset of the WebInstructSub dataset, combined with StarCoder2-Self-OSS-Instruct. Following this, we performed DPO (Direct Preference Optimization) for one epoch: using HelpSteer for the 135M and 1.7B models, and argilla/dpo-mix-7k for the 360M model. We followed the training parameters from the Zephyr-Gemma recipe in the alignment handbook, but adjusted the SFT (Supervised Fine-Tuning) learning rate to 3e-4.
-[Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
+You can load v0.1 checkpoint by specifying `revision="v0.1"` in the transformers code:
+```python
+model = AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM-1.7B-Instruct", revision="v0.1")
+```
+
+## Usage
 
-This is the SmolLM-1.7B-Instruct.
+### Local Applications
+⚡ For local applications, you can find optimized implementations of the model in MLC, GGUF and Transformers.js formats, in addition to fast in-browser demos in this collection: https://huggingface.co/collections/HuggingFaceTB/local-smollms-66c0f3b2a15b4eed7fb198d0
 
-### Generation
+We noticed that 4bit quantization degrades the quality of the 135M and 360M, so we use `q016` for MLC and ONNX/Transformers.js checkpoints for the WebGPU demos. We also suggest using temperature 0.2 and top-p 0.9.
+
+### Transformers
 ```bash
 pip install transformers
 ```
@@ -37,18 +67,35 @@ tokenizer = AutoTokenizer.from_pretrained(checkpoint)
 # for multiple GPUs install accelerate and do `model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map="auto")`
 model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)
 
-messages = [{"role": "user", "content": "List the steps to bake a chocolate cake from scratch."}]
+messages = [{"role": "user", "content": "What is the capital of France."}]
 input_text=tokenizer.apply_chat_template(messages, tokenize=False)
 print(input_text)
 inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
-outputs = model.generate(inputs, max_new_tokens=100, temperature=0.6, top_p=0.92, do_sample=True)
+outputs = model.generate(inputs, max_new_tokens=50, temperature=0.2, top_p=0.9, do_sample=True)
 print(tokenizer.decode(outputs[0]))
 ```
 
+### Chat in TRL
+You can also use the TRL CLI to chat with the model from the terminal:
+```bash
+pip install trl
+trl chat --model_name_or_path HuggingFaceTB/SmolLM-1.7B-Instruct --device cpu
+```
+
+## Limitations
+
+Additionally, the generated content may not always be factually accurate, logically consistent, or free from biases present in the training data, we invite users to leverage them as assistive tools rather than definitive sources of information. We find that they can handle general knowledge questions, creative writing and basic Python programming. But they are English only and may have difficulty with arithmetics, editing tasks and complex reasoning. For more details about the models' capabilities, please refer to our [blog post](https://huggingface.co/blog/smollm).
+
+## Training parameters
+We train the models using the [alignment-handbook](https://github.com/huggingface/alignment-handbook) with the datasets mentioned in the changelog, using these parameters v0.2:
 
-# Limitations
+- 1 epoch
+- lr 1e-3
+- cosine schedule
+- warmup ratio 0.1
+- global batch size 262k tokens
 
-While SmolLM models have been trained on a diverse dataset including educational content and synthetic texts, they have limitations. The models primarily understand and generate content in English. They can produce text on a variety of topics, but the generated content may not always be factually accurate, logically consistent, or free from biases present in the training data. These models should be used as assistive tools rather than definitive sources of information. Users should always verify important information and critically evaluate any generated content. For a more comprehensive discussion of the models' capabilities and limitations, please refer to our full blog post.
+You can find the training recipe here: https://github.com/huggingface/alignment-handbook/tree/smollm/recipes/smollm
 
 # Citation
 ```bash

all_results.json

@@ -1,22 +0,0 @@
-{
-    "epoch": 0.9872958257713249,
-    "eval_logits/chosen": 0.9550817608833313,
-    "eval_logits/rejected": 0.8631864190101624,
-    "eval_logps/chosen": -121.98506164550781,
-    "eval_logps/rejected": -149.43112182617188,
-    "eval_loss": 0.6671856641769409,
-    "eval_rewards/accuracies": 0.550000011920929,
-    "eval_rewards/chosen": -0.0465608611702919,
-    "eval_rewards/margins": 0.04670196771621704,
-    "eval_rewards/rejected": -0.09326283633708954,
-    "eval_runtime": 4.6203,
-    "eval_samples": 464,
-    "eval_samples_per_second": 100.425,
-    "eval_steps_per_second": 3.247,
-    "total_flos": 0.0,
-    "train_loss": 0.6753217893488267,
-    "train_runtime": 321.7901,
-    "train_samples": 8806,
-    "train_samples_per_second": 27.366,
-    "train_steps_per_second": 0.211
-}

eval_results.json

@@ -1,16 +0,0 @@
-{
-    "epoch": 0.9872958257713249,
-    "eval_logits/chosen": 0.9550817608833313,
-    "eval_logits/rejected": 0.8631864190101624,
-    "eval_logps/chosen": -121.98506164550781,
-    "eval_logps/rejected": -149.43112182617188,
-    "eval_loss": 0.6671856641769409,
-    "eval_rewards/accuracies": 0.550000011920929,
-    "eval_rewards/chosen": -0.0465608611702919,
-    "eval_rewards/margins": 0.04670196771621704,
-    "eval_rewards/rejected": -0.09326283633708954,
-    "eval_runtime": 4.6203,
-    "eval_samples": 464,
-    "eval_samples_per_second": 100.425,
-    "eval_steps_per_second": 3.247
-}

generation_config.json

@@ -3,5 +3,6 @@
   "bos_token_id": 1,
   "eos_token_id": 2,
   "pad_token_id": 2,
+  "max_new_tokens": 40,
   "transformers_version": "4.42.3"
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bd664d6e8e96531952daac8360d9260f4e91b287b9ea8ebebf113d25687e55f
+oid sha256:7108f45b86e4b0c9b0c8f611264b381522e093b366ec0fdf83c0e07149b0c398
 size 3422777952

onnx/model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4726fb8174a4b324f7880bf27477195617e68b9e873ca397454246dbff02a203
-size 563625
+oid sha256:5e2b596dd6bd22b4337eaf3bf4161313306dfc61d12227dab61a28a1afecd95e
+size 562234

onnx/model.onnx_data

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63c222e386f91f648ae40f9a99ebc1fe40403a00dc443a9f8b5abf0fc0a9dbc0
+oid sha256:43c2ff1f9af866f2fc88d29d9ef6c283c2c3908908bfc45e8b7acb948a4b41bc
 size 6845505536

onnx/model_bnb4.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2366a4b360655ea43277cd5f2cf8c0883b2b740dbf33b5a933da1bb668c6dd1f
-size 1309600577
+oid sha256:a632c41a9cd1bcd7116a75ce4c77c290b1e65532f001107d873c4759462a4d6b
+size 1309599186

onnx/model_fp16.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f16d1593b975e7561250d8a2c502927eb7c989401f553b67a6edcd12757a7e9
-size 806094042
+oid sha256:a38d5ebf9f8d3556de0aa6b5c7fde73bbd75b89b603f03eadc5903fb47dac346
+size 578785

onnx/model_fp16.onnx_data

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b465f6b5cb92208c0be17064b0de448cab9f29387aa3a234cc3a46d5e4ae0463
-size 2617245696
+oid sha256:bfb0a58ad2b44ced92f2cb3b5c3c6554b007d210d4fd45b090d0e78156aeb292
+size 3422752768

onnx/model_int8.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5fbaba287e62562085daca45a9119c2ce5a33422bed8ed3288393403cb1eee2
-size 1712410874
+oid sha256:c7bcbc362755972f15beb94e1d8a9e4d83e8ccf08e669d40c668061e4d712ac0
+size 1712409483

onnx/model_q4.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bbced44bc9d8d6afa0f7d37e93df11028548129349fa2924a03559366737b0a
-size 1410262529
+oid sha256:e80eb151f4c5c50e842bbf338a774436d0f4c3e3a81176d4bfc6760853d8a7a6
+size 1410261138

onnx/model_q4f16.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c389d12651b9f89e7f7efd023eba2a0f9458a819ffebdcafb26c99b0b36052d
-size 1108103905
+oid sha256:54ca7d257e47e478ecb9f1306399aa111305556eabac5485c470294df3273cf9
+size 1108085974

onnx/model_quantized.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5fbaba287e62562085daca45a9119c2ce5a33422bed8ed3288393403cb1eee2
-size 1712410874
+oid sha256:c7bcbc362755972f15beb94e1d8a9e4d83e8ccf08e669d40c668061e4d712ac0
+size 1712409483

onnx/model_uint8.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5dbd948d100c1f6893581301e12a7cdf15b327111cb45638f5b043198804788
-size 1712410958
+oid sha256:b28a17162551f5165479e10819627426cdb82d3e9dfff339a2d5f651d6a6429c
+size 1712409564

runs/Jul15_23-01-52_ip-26-0-171-168/events.out.tfevents.1721084661.ip-26-0-171-168.1537695.0

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f9e82e0a6588823112621478b95e6fc4db5d7bb0750bbc260d45a4f6d5eaa35c
-size 10294

runs/Jul15_23-01-52_ip-26-0-171-168/events.out.tfevents.1721085183.ip-26-0-171-168.1537695.1

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:27a6a863f4bf58d6242e050e5e56402f3611cf866ad775259a318e644ad89c3d
-size 815

runs/Jul15_23-02-15_ip-26-0-175-34/events.out.tfevents.1721084673.ip-26-0-175-34.1294397.0

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fee64781b1f43c757e014bacbcdc4b210d3b76ebe5fc1f4a63a28d24172c2419
-size 10293

runs/Jul15_23-02-15_ip-26-0-175-34/events.out.tfevents.1721085077.ip-26-0-175-34.1294397.1

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:038d227d30387508a3dd8bb86a277ae6c97ddc940005a81c01cc74689ec17040
-size 815

train_results.json

@@ -1,9 +0,0 @@
-{
-    "epoch": 0.9872958257713249,
-    "total_flos": 0.0,
-    "train_loss": 0.6753217893488267,
-    "train_runtime": 321.7901,
-    "train_samples": 8806,
-    "train_samples_per_second": 27.366,
-    "train_steps_per_second": 0.211
-}

trainer_state.json

@@ -1,147 +0,0 @@
-{
-  "best_metric": null,
-  "best_model_checkpoint": null,
-  "epoch": 0.9872958257713249,
-  "eval_steps": 100,
-  "global_step": 68,
-  "is_hyper_param_search": false,
-  "is_local_process_zero": true,
-  "is_world_process_zero": true,
-  "log_history": [
-    {
-      "epoch": 0.014519056261343012,
-      "grad_norm": 2.5819620488760253,
-      "learning_rate": 7.142857142857143e-07,
-      "logits/chosen": 0.902682900428772,
-      "logits/rejected": 0.8661031723022461,
-      "logps/chosen": -125.2274169921875,
-      "logps/rejected": -145.89674377441406,
-      "loss": 0.6931,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": 0.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": 0.0,
-      "step": 1
-    },
-    {
-      "epoch": 0.14519056261343014,
-      "grad_norm": 3.2242486672937045,
-      "learning_rate": 4.970219740227693e-06,
-      "logits/chosen": 0.8677890300750732,
-      "logits/rejected": 0.88941490650177,
-      "logps/chosen": -120.65478515625,
-      "logps/rejected": -149.0773468017578,
-      "loss": 0.6925,
-      "rewards/accuracies": 0.4097222089767456,
-      "rewards/chosen": 0.0005513947107829154,
-      "rewards/margins": -0.0005504629807546735,
-      "rewards/rejected": 0.001101857633329928,
-      "step": 10
-    },
-    {
-      "epoch": 0.29038112522686027,
-      "grad_norm": 4.0763010346326904,
-      "learning_rate": 4.460299516441777e-06,
-      "logits/chosen": 1.0478929281234741,
-      "logits/rejected": 1.0530675649642944,
-      "logps/chosen": -115.20359802246094,
-      "logps/rejected": -142.4711456298828,
-      "loss": 0.6887,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": -0.007147049065679312,
-      "rewards/margins": 0.007046631071716547,
-      "rewards/rejected": -0.014193681068718433,
-      "step": 20
-    },
-    {
-      "epoch": 0.4355716878402904,
-      "grad_norm": 2.868995154924774,
-      "learning_rate": 3.441819734087963e-06,
-      "logits/chosen": 1.0493364334106445,
-      "logits/rejected": 1.0137263536453247,
-      "logps/chosen": -115.47831726074219,
-      "logps/rejected": -145.4583282470703,
-      "loss": 0.681,
-      "rewards/accuracies": 0.53125,
-      "rewards/chosen": -0.013122054748237133,
-      "rewards/margins": 0.018696457147598267,
-      "rewards/rejected": -0.031818509101867676,
-      "step": 30
-    },
-    {
-      "epoch": 0.5807622504537205,
-      "grad_norm": 2.4635583297317165,
-      "learning_rate": 2.1790041121336223e-06,
-      "logits/chosen": 1.2130175828933716,
-      "logits/rejected": 1.1768478155136108,
-      "logps/chosen": -154.19398498535156,
-      "logps/rejected": -181.88272094726562,
-      "loss": 0.6762,
-      "rewards/accuracies": 0.581250011920929,
-      "rewards/chosen": -0.04483573138713837,
-      "rewards/margins": 0.041850604116916656,
-      "rewards/rejected": -0.08668633550405502,
-      "step": 40
-    },
-    {
-      "epoch": 0.7259528130671506,
-      "grad_norm": 3.188683944287027,
-      "learning_rate": 9.994642986290797e-07,
-      "logits/chosen": 1.3518016338348389,
-      "logits/rejected": 1.2714142799377441,
-      "logps/chosen": -131.69985961914062,
-      "logps/rejected": -161.8300018310547,
-      "loss": 0.6622,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.011605720035731792,
-      "rewards/margins": 0.06830872595310211,
-      "rewards/rejected": -0.07991443574428558,
-      "step": 50
-    },
-    {
-      "epoch": 0.8711433756805808,
-      "grad_norm": 3.3445679325076476,
-      "learning_rate": 2.092077387824884e-07,
-      "logits/chosen": 1.367594838142395,
-      "logits/rejected": 1.2992292642593384,
-      "logps/chosen": -121.10174560546875,
-      "logps/rejected": -139.6844940185547,
-      "loss": 0.6612,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": -0.017885591834783554,
-      "rewards/margins": 0.07065734267234802,
-      "rewards/rejected": -0.08854293823242188,
-      "step": 60
-    },
-    {
-      "epoch": 0.9872958257713249,
-      "step": 68,
-      "total_flos": 0.0,
-      "train_loss": 0.6753217893488267,
-      "train_runtime": 321.7901,
-      "train_samples_per_second": 27.366,
-      "train_steps_per_second": 0.211
-    }
-  ],
-  "logging_steps": 10,
-  "max_steps": 68,
-  "num_input_tokens_seen": 0,
-  "num_train_epochs": 1,
-  "save_steps": 500,
-  "stateful_callbacks": {
-    "TrainerControl": {
-      "args": {
-        "should_epoch_stop": false,
-        "should_evaluate": false,
-        "should_log": false,
-        "should_save": false,
-        "should_training_stop": false
-      },
-      "attributes": {}
-    }
-  },
-  "total_flos": 0.0,
-  "train_batch_size": 2,
-  "trial_name": null,
-  "trial_params": null
-}

training_args.bin

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c619b9db98ea959618a4839c7d36e8399c8ce4dac6b0fc49c96d5e1d3b9e1fc4
-size 6520