awq_w3g128_L32_3b.yml

base:
    seed: &seed 42
model:
    type: Llama
    path: meta-llama/Llama-3.2-3B
    torch_dtype: auto
calib:
    name: pileval
    download: False
    path: ../cache/data/calib/pileval
    n_samples: 128
    bs: -1
    seq_len: 512
    preproc: pileval_awq
    seed: *seed
eval:
    eval_pos: [fake_quant]
    name: [wikitext2, c4]
    download: False
    path: ../cache/data/eval
    bs: 1
    inference_per_block: False
    seq_len: 2048
    tasks: piqa,arc_easy,arc_challenge,hellaswag,winogrande
quant:
    method: Awq
    weight:
        bit: 3
        symmetric: False
        granularity: per_group
        group_size: 128
        calib_algo: learnable
    special:
        trans: True
        weight_clip: True
        save_scale: True
        clip_version: v1
        scale_path: ../cache/activations/L32_3b/awq_w3g128
        save_clip: False
        clip_path: ../cache/activations/L32_3b/awq_w3g128
save:
    save_fp: False
    save_lightllm: False
    save_vllm: False
    save_trans: False