--- tags: - int4 - vllm language: - en - de - fr - it - pt - hi - es - th pipeline_tag: text-generation license: llama3.1 base_model: meta-llama/Meta-Llama-3.1-70B-Instruct --- # Meta-Llama-3.1-70B-Instruct-quantized.w4a16 ## Model Overview - **Model Architecture:** Meta-Llama-3 - **Input:** Text - **Output:** Text - **Model Optimizations:** - **Weight quantization:** INT4 - **Intended Use Cases:** Intended for commercial and research use in English. Similarly to [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-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:** 7/26/2024 - **Version:** 1.0 - **License(s):** Llama3.1 - **Model Developers:** Neural Magic This model is a quantized version of [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct). It was evaluated on a several tasks to assess the its quality in comparison to the unquatized model, including multiple-choice, math reasoning, and open-ended text generation. Meta-Llama-3.1-70B-Instruct-quantized.w4a16 achieves 100.0% recovery for the Arena-Hard evaluation, 99.4% for OpenLLM v1 (using Meta's prompting when available), 97.4% for OpenLLM v2, 101.0% for HumanEval pass@1, and 99.2% for HumanEval+ pass@1. ### Model Optimizations This model was obtained by quantizing the weights of [Meta-Llama-3.1-70B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-70B-Instruct) to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%. Only the weights of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the INT4 and floating point representations of the quantized weights. The [GPTQ](https://arxiv.org/abs/2210.17323) algorithm is applied for quantization, as implemented in the [AutoGPTQ](https://github.com/AutoGPTQ/AutoGPTQ) library. GPTQ used a 1% damping factor and 512 sequences of 8,192 random tokens. ## Deployment 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/Meta-Llama-3.1-70B-Instruct-quantized.w4a16" number_gpus = 1 max_model_len = 8192 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, tensor_parallel_size=number_gpus, max_model_len=max_model_len) 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 applying the [AutoGPTQ](https://github.com/AutoGPTQ/AutoGPTQ) library as presented in the code snipet below. Although AutoGPTQ 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 AutoGPTQ. ```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 = "meta-llama/Meta-Llama-3.1-70B-Instruct" num_samples = 512 max_seq_len = 8192 tokenizer = AutoTokenizer.from_pretrained(model_id) preprocess_fn = lambda example: {"text": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n{text}".format_map(example)} dataset_name = "neuralmagic/LLM_compression_calibration" dataset = load_dataset(dataset_name, split="train") ds = dataset.shuffle().select(range(num_samples)) ds = ds.map(preprocess_fn) recipe = GPTQModifier( targets="Linear", scheme="W4A16", ignore=["lm_head"], dampening_frac=0.01, ) 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("Meta-Llama-3.1-70B-Instruct-quantized.w4a16") ``` ## Evaluation This model was evaluated on the well-known Arena-Hard, OpenLLM v1, OpenLLM v2, HumanEval, and HumanEval+ benchmarks. In all cases, model outputs were generated with the [vLLM](https://docs.vllm.ai/en/stable/) engine. Arena-Hard evaluations were conducted using the [Arena-Hard-Auto](https://github.com/lmarena/arena-hard-auto) repository. The model generated a single answer for each prompt form Arena-Hard, and each answer was judged twice by GPT-4. We report below the scores obtained in each judgement and the average. OpenLLM v1 and v2 evaluations were conducted using Neural Magic's fork of [lm-evaluation-harness](https://github.com/neuralmagic/lm-evaluation-harness/tree/llama_3.1_instruct) (branch llama_3.1_instruct). This version of the lm-evaluation-harness includes versions of MMLU, ARC-Challenge and GSM-8K that match the prompting style of [Meta-Llama-3.1-Instruct-evals](https://huggingface.co/datasets/meta-llama/Meta-Llama-3.1-70B-Instruct-evals) and a few fixes to OpenLLM v2 tasks. HumanEval and HumanEval+ evaluations were conducted using Neural Magic's fork of the [EvalPlus](https://github.com/neuralmagic/evalplus) repository. Detailed model outputs are available as HuggingFace datasets for [Arena-Hard](https://huggingface.co/datasets/neuralmagic/quantized-llama-3.1-arena-hard-evals), [OpenLLM v2](https://huggingface.co/datasets/neuralmagic/quantized-llama-3.1-leaderboard-v2-evals), and [HumanEval](https://huggingface.co/datasets/neuralmagic/quantized-llama-3.1-humaneval-evals). **Note:** Results have been updated after Meta modified the chat template. ### Accuracy
Benchmark Meta-Llama-3.1-70B-Instruct Meta-Llama-3.1-70B-Instruct-quantized.w4a16 (this model) Recovery
Arena Hard 57.0 (55.8 / 58.2) 57.0 (57.1 / 56.8) 100.0%
OpenLLM v1
MMLU (5-shot) 83.9 83.6 99.5%
MMLU (CoT, 0-shot) 86.2 85.6 99.2%
ARC Challenge (0-shot) 93.3 92.8 99.5%
GSM-8K (CoT, 8-shot, strict-match) 95.4 94.4 99.0%
Hellaswag (10-shot) 86.7 86.3 99.5%
Winogrande (5-shot) 85.3 85.5 100.2%
TruthfulQA (0-shot, mc2) 60.7 59.8 98.6%
Average 84.50 83.98 99.4%
OpenLLM v2
MMLU-Pro (5-shot) 48.1 47.3 98.2%
IFEval (0-shot) 86.4 85.7 99.2%
BBH (3-shot) 55.8 55.0 98.6%
Math-|v|-5 (4-shot) 26.1 24.4 93.5%
GPQA (0-shot) 15.4 13.9 89.9%
MuSR (0-shot) 18.2 17.3 95.0%
Average 41.7 40.6 97.4%
Coding
HumanEval pass@1 79.7 80.5 101.0%
HumanEval+ pass@1 74.8 74.2 99.2%
### Reproduction The results were obtained using the following commands: #### MMLU ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,max_model_len=3850,max_gen_toks=10,tensor_parallel_size=1 \ --tasks mmlu_llama_3.1_instruct \ --fewshot_as_multiturn \ --apply_chat_template \ --num_fewshot 5 \ --batch_size auto ``` #### MMLU-CoT ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,max_model_len=4064,max_gen_toks=1024,tensor_parallel_size=1 \ --tasks mmlu_cot_0shot_llama_3.1_instruct \ --apply_chat_template \ --num_fewshot 0 \ --batch_size auto ``` #### ARC-Challenge ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,max_model_len=3940,max_gen_toks=100,tensor_parallel_size=1 \ --tasks arc_challenge_llama_3.1_instruct \ --apply_chat_template \ --num_fewshot 0 \ --batch_size auto ``` #### GSM-8K ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,max_model_len=4096,max_gen_toks=1024,tensor_parallel_size=1 \ --tasks gsm8k_cot_llama_3.1_instruct \ --fewshot_as_multiturn \ --apply_chat_template \ --num_fewshot 8 \ --batch_size auto ``` #### Hellaswag ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \ --tasks hellaswag \ --num_fewshot 10 \ --batch_size auto ``` #### Winogrande ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \ --tasks winogrande \ --num_fewshot 5 \ --batch_size auto ``` #### TruthfulQA ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=1 \ --tasks truthfulqa \ --num_fewshot 0 \ --batch_size auto ``` #### OpenLLM v2 ``` lm_eval \ --model vllm \ --model_args pretrained="neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16",dtype=auto,max_model_len=4096,tensor_parallel_size=1,enable_chunked_prefill=True \ --apply_chat_template \ --fewshot_as_multiturn \ --tasks leaderboard \ --batch_size auto ``` #### HumanEval and HumanEval+ ##### Generation ``` python3 codegen/generate.py \ --model neuralmagic/Meta-Llama-3.1-70B-Instruct-quantized.w4a16 \ --bs 16 \ --temperature 0.2 \ --n_samples 50 \ --root "." \ --dataset humaneval ``` ##### Sanitization ``` python3 evalplus/sanitize.py \ humaneval/neuralmagic--Meta-Llama-3.1-70B-Instruct-quantized.w4a16_vllm_temp_0.2 ``` ##### Evaluation ``` evalplus.evaluate \ --dataset humaneval \ --samples humaneval/neuralmagic--Meta-Llama-3.1-70B-Instruct-quantized.w4a16_vllm_temp_0.2-sanitized ```