Update README.md

README.md

@@ -17,13 +17,13 @@ license_link: https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/
   - **Activation quantization:** FP8
 - **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct), this models is intended for assistant-like chat.
 - **Out-of-scope:** Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
-- **Release Date:** 6/29/2024
-- **Version:** 1.0
+- **Release Date:** 8/11/2024
+- **Version:** 1.1
 - **License(s):** [mit](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/LICENSE)
 - **Model Developers:** Neural Magic
 
-Quantized version of [Phi-3-mini-128k-instruct](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct).
-It achieves an average score of 68.99 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 69.13.
+Quantized version of [Phi-3-mini-128k-instruct](https://huggingface.co/microsoft/Phi-3-mini-128k-instruct), with the new configuration files.
+It achieves an average score of 69.42 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 69.69.
 
 ### Model Optimizations
 
@@ -68,32 +68,93 @@ vLLM aslo supports OpenAI-compatible serving. See the [documentation](https://do
 
 ## Creation
 
-This model was created by applying [AutoFP8 with calibration samples from ultrachat](https://github.com/neuralmagic/AutoFP8/blob/147fa4d9e1a90ef8a93f96fc7d9c33056ddc017a/example_dataset.py), as presented in the code snipet below.
-Although AutoFP8 was used for this particular model, Neural Magic is transitioning to using [llm-compressor](https://github.com/vllm-project/llm-compressor) which supports several quantization schemes and models not supported by AutoFP8.
+This model was created by applying [LLM Compressor with calibration samples from UltraChat](https://github.com/vllm-project/llm-compressor/blob/sa/big_model_support/examples/big_model_offloading/big_model_w8a8_calibrate.py), as presented in the code snipet below.
 
 ```python
+import torch
 from datasets import load_dataset
 from transformers import AutoTokenizer
 
-from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
-
-pretrained_model_dir = "microsoft/Phi-3-mini-128k-instruct"
-quantized_model_dir = "Phi-3-mini-128k-instruct-FP8"
-
-tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True, model_max_length=4096)
-tokenizer.pad_token = tokenizer.eos_token
-
-ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(range(512))
-examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
-examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
+from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
+from llmcompressor.transformers.compression.helpers import (
+    calculate_offload_device_map,
+    custom_offload_device_map,
+)
 
-quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="static")
+recipe = """
+quant_stage:
+    quant_modifiers:
+        QuantizationModifier:
+            ignore: ["lm_head"]
+            config_groups:
+                group_0:
+                    weights:
+                        num_bits: 8
+                        type: float
+                        strategy: tensor
+                        dynamic: false
+                        symmetric: true
+                    input_activations:
+                        num_bits: 8
+                        type: float
+                        strategy: tensor
+                        dynamic: false
+                        symmetric: true
+                    targets: ["Linear"]
+"""
+
+model_stub = "microsoft/Phi-3-mini-128k-instruct"
+model_name = model_stub.split("/")[-1]
+
+device_map = calculate_offload_device_map(
+    model_stub, reserve_for_hessians=False, num_gpus=1, torch_dtype=torch.float16
+)
 
-model = AutoFP8ForCausalLM.from_pretrained(
-    pretrained_model_dir, quantize_config=quantize_config
+model = SparseAutoModelForCausalLM.from_pretrained(
+    model_stub, torch_dtype=torch.float16, device_map=device_map
+)
+tokenizer = AutoTokenizer.from_pretrained(model_stub)
+
+output_dir = f"./{model_name}-FP8"
+
+DATASET_ID = "HuggingFaceH4/ultrachat_200k"
+DATASET_SPLIT = "train_sft"
+NUM_CALIBRATION_SAMPLES = 512
+MAX_SEQUENCE_LENGTH = 4096
+
+ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
+ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))
+
+def preprocess(example):
+    return {
+        "text": tokenizer.apply_chat_template(
+            example["messages"],
+            tokenize=False,
+        )
+    }
+
+ds = ds.map(preprocess)
+
+def tokenize(sample):
+    return tokenizer(
+        sample["text"],
+        padding=False,
+        max_length=MAX_SEQUENCE_LENGTH,
+        truncation=True,
+        add_special_tokens=False,
+    )
+
+ds = ds.map(tokenize, remove_columns=ds.column_names)
+
+oneshot(
+    model=model,
+    output_dir=output_dir,
+    dataset=ds,
+    recipe=recipe,
+    max_seq_length=MAX_SEQUENCE_LENGTH,
+    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
+    save_compressed=True,
 )
-model.quantize(examples)
-model.save_quantized(quantized_model_dir)
 ```
 
 ## Evaluation
@@ -124,71 +185,71 @@ lm_eval \
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>68.10
+   <td>69.33
    </td>
-   <td>67.93
+   <td>68.90
    </td>
-   <td>99.75%
+   <td>99.38%
    </td>
   </tr>
   <tr>
    <td>ARC Challenge (25-shot)
    </td>
-   <td>63.65
+   <td>63.05
    </td>
-   <td>64.24
+   <td>63.05
    </td>
-   <td>100.9%
+   <td>100.0%
    </td>
   </tr>
   <tr>
    <td>GSM-8K (5-shot, strict-match)
    </td>
-   <td>75.59
+   <td>76.95
    </td>
-   <td>74.37
+   <td>76.27
    </td>
-   <td>98.38%
+   <td>99.12%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>79.76
+   <td>79.58
    </td>
-   <td>79.79
+   <td>79.36
    </td>
-   <td>100.0%
+   <td>99.72%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
-   <td>73.72
+   <td>74.82
    </td>
-   <td>74.11
+   <td>74.59
    </td>
-   <td>100.5%
+   <td>99.69%
    </td>
   </tr>
   <tr>
    <td>TruthfulQA (0-shot)
    </td>
-   <td>53.97
+   <td>54.41
    </td>
-   <td>53.50
+   <td>54.36
    </td>
-   <td>99.12%
+   <td>99.91%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>69.13</strong>
+   <td><strong>69.69</strong>
    </td>
-   <td><strong>68.99</strong>
+   <td><strong>69.42</strong>
    </td>
-   <td><strong>99.80%</strong>
+   <td><strong>99.61%</strong>
    </td>
   </tr>
 </table>