--- language: - en pipeline_tag: text-generation license: mit --- # Phi-3-medium-128k-instruct-quantized.w8a8 ## Model Overview - **Model Architecture:** Phi-3 - **Input:** Text - **Output:** Text - **Model Optimizations:** - **Activation quantization:** INT8 - **Weight quantization:** INT8 - **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-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:** 7/11/2024 - **Version:** 1.0 - **License(s):** [MIT](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/mit.md) - **Model Developers:** Neural Magic Quantized version of [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct), a 14 billion-parameter open model trained using the Phi-3 datasets. It achieves an average score of 73.90 on the [OpenLLM](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) benchmark (version 1), whereas the unquantized model achieves 74.10. ### Model Optimizations This model was obtained by quantizing the weights of [Phi-3-medium-128k-instruct](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct) to INT8 data type. This optimization reduces the number of bits used to represent weights and activations from 16 to 8, reducing GPU memory requirements (by approximately 50%) and increasing matrix-multiply compute throughput (by approximately 2x). Weight quantization also reduces disk size requirements by approximately 50%. Only weights and activations of the linear operators within transformers blocks are quantized. Weights are quantized with a symmetric static per-channel scheme, where a fixed linear scaling factor is applied between INT8 and floating point representations for each output channel dimension. Activations are quantized with a symmetric dynamic per-token scheme, computing a linear scaling factor at runtime for each token between INT8 and floating point representations. Linear scaling factors are computed via by minimizing the mean squarred error (MSE). The [SmoothQuant](https://arxiv.org/abs/2211.10438) algorithm is used to alleviate outliers in the activations, whereas rhe [GPTQ](https://arxiv.org/abs/2210.17323) algorithm is applied for quantization. Both algorithms are implemented in the [llm-compressor](https://github.com/vllm-project/llm-compressor) library. GPTQ used a 1% damping factor and 512 sequences sequences taken from Neural Magic's [LLM compression calibration dataset](https://huggingface.co/datasets/neuralmagic/LLM_compression_calibration). ## 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 (using 2 GPUs). ```python from vllm import LLM, SamplingParams from transformers import AutoTokenizer model_id = "neuralmagic/Phi-3-medium-128k-instruct-quantized.w8a8" number_gpus = 2 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?"}, ] prompts = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) llm = LLM(model=model_id, trust_remote_code=True, max_model_len=8196, tensor_parallel_size=number_gpus) outputs = llm.generate(prompts, sampling_params) generated_text = outputs[0].outputs[0].text print(generated_text) ``` vLLM aslo supports OpenAI-compatible serving. See the [documentation](https://docs.vllm.ai/en/latest/) for more details. ## Creation This model was created by using the [llm-compressor](https://github.com/vllm-project/llm-compressor) library as presented in the code snipet below. ```python from transformers import AutoTokenizer from datasets import Dataset from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot from llmcompressor.modifiers.quantization import GPTQModifier import random model_id = "microsoft/Phi-3-medium-128k-instruct" num_samples = 512 max_seq_len = 8192 tokenizer = AutoTokenizer.from_pretrained(model_id) def preprocess_fn(example): return {"text": tokenizer.apply_chat_template(example["messages"], add_generation_prompt=False, tokenize=False)} ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train") ds = ds.shuffle().select(range(num_samples)) ds = ds.map(preprocess_fn) recipe = [ SmoothQuantModifier( smoothing_strength=0.8, mappings=[ [["re:.*qkv_proj"], "re:.*input_layernorm"], [["re:.*gate_up_proj"], "re:.*post_attention_layernorm"], ], ), GPTQModifier( sequential=True, targets="Linear", scheme="W8A8", ignore=["lm_head"], dampening_frac=0.01, observer="mse", ) ] model = SparseAutoModelForCausalLM.from_pretrained( model_id, device_map="auto", trust_remote_code=True, ) oneshot( model=model, dataset=ds, recipe=recipe, max_seq_length=max_seq_len, num_calibration_samples=num_samples, ) model.save_pretrained("Phi-3-medium-128k-instruct-quantized.w8a8") ``` ## 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 (using 2 GPUs): ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Phi-3-medium-128k-instruct-quantized.w8a8",dtype=auto,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096,tensor_parallel_size=2 \ --tasks openllm \ --batch_size auto ``` ### Accuracy #### Open LLM Leaderboard evaluation scores
Benchmark Phi-3-medium-128k-instruct Phi-3-medium-128k-instruct-quantized.w8a8 (this model) Recovery
MMLU (5-shot) 76.69 76.74 100.1%
ARC Challenge (25-shot) 69.45 69.37 99.9%
GSM-8K (5-shot, strict-match) 85.22 84.15 98.7%
Hellaswag (10-shot) 85.10 84.76 99.6%
Winogrande (5-shot) 73.56 73.80 100.3%
TruthfulQA (0-shot) 54.57 54.57 100.0%
Average 74.10 73.90 99.7%