yujiepan/awq-model-zoo

Here are some pre-computed awq information (scales & clips) used in llm-awq.

Scripts

• Install the forked llm-awq at https://github.com/yujiepan-work/llm-awq/tree/a41a08e79d8eb3d6335485b3625410af22a74426. Note: works with transformers==4.35.2

• Generating awq-info.pt:

python do_awq.py --model_id mistralai/Mistral-7B-v0.1 --w_bit 8 --q_group_size 128 --dump_awq ./awq-info.pt

• Load a quantized model: You can use the offical repo to get a fake/real quantized model. Alternatively, you can load a fake-quantized model:

from do_awq import FakeAWQModel
FakeAWQModel.from_pretrained('mistralai/Mistral-7B-v0.1', awq_meta_path='./awq-info.pt',  output_folder='./tmp/')

Note: the code is not in good shape.

Related links

• https://huggingface.co/datasets/mit-han-lab/awq-model-zoo