Evaluated on lm-evaluation-harness "wikitext" task
Wikitext PPL does not guarantee actual accuracy, but helps to check the disortion after quantization.

Usage

python

from awq import AutoAWQForCausalLM
model = AutoAWQForCausalLM.from_quantized('yujiepan/Meta-Llama-3-8B-awq-w4g64')

Codes

python

from awq import AutoAWQForCausalLM
from transformers import AutoTokenizer

model_path = "meta-llama/Meta-Llama-3-8B"
quant_config = {"zero_point": True, "q_group_size": 64, "w_bit": 4, "version": "GEMM"}

model = AutoAWQForCausalLM.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model.quantize(tokenizer, quant_config=quant_config)