Update README.md

README.md

@@ -17,13 +17,13 @@ license_link: https://huggingface.co/microsoft/Phi-3-medium-128k-instruct/resolv
   - **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:** 8/12/2024
+- **Version:** 1.1
 - **License(s):** [mit](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct/resolve/main/LICENSE)
 - **Model Developers:** Neural Magic
 
-Quantized version of [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct).
-It achieves an average score of 73.20 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 73.47.
+Quantized version of [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct), with the new configuration files.
+It achieves an average score of 73.65 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 73.95.
 
 ### Model Optimizations
 
@@ -68,50 +68,94 @@ vLLM also 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.
+Importantly, the "rope_scaling" type in config.json was manually changed from "longrope" to "su" following quantization.
 
 ```python
+import torch
 from datasets import load_dataset
 from transformers import AutoTokenizer
-import numpy as np
-import torch
-
-from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
-
-MODEL_DIR = "microsoft/Phi-3-medium-128k-instruct"
-final_model_dir = MODEL_DIR.split("/")[-1]
-
-CONTEXT_LENGTH = 4096
-NUM_SAMPLES = 512
-NUM_REPEATS = 10
-
-pretrained_model_dir = MODEL_DIR
-tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True, model_max_length=CONTEXT_LENGTH)
-tokenizer.pad_token = tokenizer.eos_token
-
-tokenizer_num_tokens = len(list(tokenizer.get_vocab().values()))
-total_token_samples = NUM_REPEATS * tokenizer_num_tokens
-num_random_samp = -(-total_token_samples // CONTEXT_LENGTH)
 
-input_ids = np.tile(np.arange(tokenizer_num_tokens), NUM_REPEATS + 1)[:num_random_samp * CONTEXT_LENGTH]
-np.random.shuffle(input_ids)
-input_ids = input_ids.reshape(num_random_samp, CONTEXT_LENGTH)
-input_ids = torch.tensor(input_ids, dtype=torch.int64).to("cuda")
-
-quantize_config = BaseQuantizeConfig(
-    quant_method="fp8",
-    activation_scheme="static",
+from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
+from llmcompressor.transformers.compression.helpers import (
+    calculate_offload_device_map,
+    custom_offload_device_map,
 )
 
-examples = input_ids
-
-model = AutoFP8ForCausalLM.from_pretrained(pretrained_model_dir, quantize_config=quantize_config)
-
-model.quantize(examples)
+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-medium-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
+)
 
-quantized_model_dir = f"{final_model_dir}-FP8"
-model.save_quantized(quantized_model_dir)
+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,
+)
 ```
 
 ## Evaluation
@@ -120,7 +164,7 @@ The model was evaluated on the [OpenLLM](https://huggingface.co/spaces/open-llm-
 ```
 lm_eval \
   --model vllm \
-  --model_args pretrained="neuralmagic/Phi-3-medium-128k-instruct-FP8",dtype=auto,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096 \
+  --model_args pretrained="neuralmagic/Phi-3-medium-128k-instruct-FP8",dtype=auto,gpu_memory_utilization=0.7,add_bos_token=True,max_model_len=4096 \
   --tasks openllm \
   --batch_size auto
 ```
@@ -142,71 +186,71 @@ lm_eval \
   <tr>
    <td>MMLU (5-shot)
    </td>
-   <td>75.73
+   <td>76.53
    </td>
-   <td>75.58
+   <td>76.66
    </td>
-   <td>99.80%
+   <td>100.1%
    </td>
   </tr>
   <tr>
    <td>ARC Challenge (25-shot)
    </td>
-   <td>67.66
+   <td>68.17
    </td>
-   <td>66.89
+   <td>67.06
    </td>
-   <td>98.86%
+   <td>98.37%
    </td>
   </tr>
   <tr>
    <td>GSM-8K (5-shot, strict-match)
    </td>
-   <td>84.00
+   <td>84.46
    </td>
-   <td>82.71
+   <td>84.31
    </td>
-   <td>98.46%
+   <td>99.82%
    </td>
   </tr>
   <tr>
    <td>Hellaswag (10-shot)
    </td>
-   <td>84.37
+   <td>84.77
    </td>
-   <td>84.14
+   <td>84.63
    </td>
-   <td>99.72%
+   <td>99.83%
    </td>
   </tr>
   <tr>
    <td>Winogrande (5-shot)
    </td>
-   <td>75.53
+   <td>75.22
    </td>
-   <td>74.35
+   <td>74.51
    </td>
-   <td>98.43%
+   <td>99.06%
    </td>
   </tr>
   <tr>
    <td>TruthfulQA (0-shot)
    </td>
-   <td>53.53
+   <td>54.52
    </td>
-   <td>55.52
+   <td>54.71
    </td>
-   <td>103.7%
+   <td>100.35%
    </td>
   </tr>
   <tr>
    <td><strong>Average</strong>
    </td>
-   <td><strong>73.47</strong>
+   <td><strong>73.95</strong>
    </td>
-   <td><strong>73.20</strong>
+   <td><strong>73.65</strong>
    </td>
-   <td><strong>99.63%</strong>
+   <td><strong>99.60%</strong>
    </td>
   </tr>
 </table>