--- tags: - fp8 - vllm license: mit license_link: https://huggingface.co/microsoft/Phi-3-medium-128k-instruct/resolve/main/LICENSE --- # Phi-3-medium-128k-instruct-FP8 ## Model Overview - **Model Architecture:** Phi-3 - **Input:** Text - **Output:** Text - **Model Optimizations:** - **Weight quantization:** FP8 - **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:** 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), 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 This model was obtained by quantizing the weights and activations of [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct) to FP8 data type, ready for inference with vLLM >= 0.5.1. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%. Only the weights and activations of the linear operators within transformers blocks are quantized. Symmetric per-tensor quantization is applied, in which a single linear scaling maps the FP8 representations of the quantized weights and activations. [AutoFP8](https://github.com/neuralmagic/AutoFP8) is used for quantization with 10 repeats of every token in random order. ## Deployment ### Use with vLLM This model can be deployed efficiently using the [vLLM](https://docs.vllm.ai/en/latest/) backend, as shown in the example below. ```python from vllm import LLM, SamplingParams from transformers import AutoTokenizer model_id = "neuralmagic/Phi-3-medium-128k-instruct-FP8" sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256) tokenizer = AutoTokenizer.from_pretrained(model_id) messages = [ {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"}, {"role": "user", "content": "Who are you? Remember to respond in pirate speak!"}, ] prompts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) llm = LLM(model=model_id) outputs = llm.generate(prompts, sampling_params) generated_text = outputs[0].outputs[0].text print(generated_text) ``` vLLM also supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details. ## Creation 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 from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot from llmcompressor.transformers.compression.helpers import ( calculate_offload_device_map, custom_offload_device_map, ) 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 ) 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 The model was evaluated on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) leaderboard tasks (version 1) with the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness/tree/383bbd54bc621086e05aa1b030d8d4d5635b25e6) (commit 383bbd54bc621086e05aa1b030d8d4d5635b25e6) and the [vLLM](https://docs.vllm.ai/en/stable/) engine, using the following command: ``` lm_eval \ --model vllm \ --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 ``` ### Accuracy #### Open LLM Leaderboard evaluation scores
Benchmark Phi-3-medium-128k-instruct Phi-3-medium-128k-instruct-FP8(this model) Recovery
MMLU (5-shot) 76.53 76.66 100.1%
ARC Challenge (25-shot) 68.17 67.06 98.37%
GSM-8K (5-shot, strict-match) 84.46 84.31 99.82%
Hellaswag (10-shot) 84.77 84.63 99.83%
Winogrande (5-shot) 75.22 74.51 99.06%
TruthfulQA (0-shot) 54.52 54.71 100.35%
Average 73.95 73.65 99.60%