Meta-Llama-3.1-8B-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-3.1-8B-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: 7/26/2024
• Version: 1.0
• License(s): Llama3.1
• Model Developers: Neural Magic

This model is a quantized version of Meta-Llama-3.1-8B-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-3.1-8B-Instruct-quantized.w4a16 achieves 93.0% recovery for the Arena-Hard evaluation, 98.9% for OpenLLM v1 (using Meta's prompting when available), 96.1% for OpenLLM v2, 99.7% for HumanEval pass@1, and 97.4% for HumanEval+ pass@1.

Model Optimizations

This model was obtained by quantizing the weights of Meta-Llama-3.1-8B-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 INT8 and floating point representations of the quantized weights. AutoGPTQ is used for quantization with 10% damping factor and 768 sequences taken from Neural Magic's LLM compression calibration dataset.

Deployment

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "neuralmagic/Meta-Llama-3.1-8B-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 for more details.

Creation

This model was created by applying the 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 which supports several quantization schemes and models not supported by AutoGPTQ.

from transformers import AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
from datasets import load_dataset

model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"

num_samples = 756
max_seq_len = 4064

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)

examples = [tokenizer(example["text"], padding=False, max_length=max_seq_len, truncation=True) for example in ds]
    
quantize_config = BaseQuantizeConfig(
  bits=4,
  group_size=128,
  desc_act=True,
  model_file_base_name="model",
  damp_percent=0.1,
)

model = AutoGPTQForCausalLM.from_pretrained(
  model_id,
  quantize_config,
  device_map="auto",
)

model.quantize(examples)
model.save_pretrained("Meta-Llama-3.1-8B-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 engine.

Arena-Hard evaluations were conducted using the 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 (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 and a few fixes to OpenLLM v2 tasks.

HumanEval and HumanEval+ evaluations were conducted using Neural Magic's fork of the EvalPlus repository.

Detailed model outputs are available as HuggingFace datasets for Arena-Hard, OpenLLM v2, and HumanEval.

Note: Results have been updated after Meta modified the chat template.

Accuracy

Benchmark	Meta-Llama-3.1-8B-Instruct	Meta-Llama-3.1-8B-Instruct-quantized.w4a16 (this model)	Recovery
Arena Hard	25.8 (25.1 / 26.5)	24.0 (23.4 / 24.6)	93.0%
OpenLLM v1
MMLU (5-shot)	68.3	66.9	97.9%
MMLU (CoT, 0-shot)	72.8	71.1	97.6%
ARC Challenge (0-shot)	81.4	80.2	98.0%
GSM-8K (CoT, 8-shot, strict-match)	82.8	82.9	100.2%
Hellaswag (10-shot)	80.5	79.9	99.3%
Winogrande (5-shot)	78.1	78.0	99.9%
TruthfulQA (0-shot, mc2)	54.5	52.8	96.9%
Average	74.3	73.5	98.9%
OpenLLM v2
MMLU-Pro (5-shot)	30.8	28.8	93.6%
IFEval (0-shot)	77.9	76.3	98.0%
BBH (3-shot)	30.1	28.9	96.1%
Math-|v|-5 (4-shot)	15.7	14.8	94.4%
GPQA (0-shot)	3.7	4.0	109.8%
MuSR (0-shot)	7.6	6.3	83.2%
Average	27.6	26.5	96.1%
Coding
HumanEval pass@1	67.3	67.1	99.7%
HumanEval+ pass@1	60.7	59.1	97.4%

Reproduction

The results were obtained using the following commands:

MMLU

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Meta-Llama-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-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-3.1-8B-Instruct-quantized.w4a16 \
  --bs 16 \
  --temperature 0.2 \
  --n_samples 50 \
  --root "." \
  --dataset humaneval

Sanitization

python3 evalplus/sanitize.py \
  humaneval/neuralmagic--Meta-Llama-3.1-8B-Instruct-quantized.w4a16_vllm_temp_0.2

Evaluation

evalplus.evaluate \
  --dataset humaneval \
  --samples humaneval/neuralmagic--Meta-Llama-3.1-8B-Instruct-quantized.w4a16_vllm_temp_0.2-sanitized