tesseraq_w4g128_L32_3b.yml

base:
    seed: &seed 42
model:
    type: Llama
    path: meta-llama/Llama-3.2-3B
    torch_dtype: auto
calib:
    name: wikitext2
    download: False
    n_samples: 512
    path: ../cache/data/calib/wikitext2
    bs: 1
    seq_len: 2048
    preproc: wikitext2_gptq
    seed: *seed
eval:
    eval_pos: [fake_quant]
    name: [wikitext2, c4]
    download: False
    path: ../cache/data/eval
    bs: 1
    seq_len: 2048
    inference_per_block: False
    tasks: piqa,arc_easy,arc_challenge,hellaswag,winogrande
quant:
    method: TesseraQ
    weight:
        bit: 4
        symmetric: False
        granularity: per_group
        group_size: 128
        calib_algo: minmax
    special:
        lr: 0.001
        iterations: 250
        wd: 0.0
        batch_size: 4
        deactive_amp: False
        aug_loss: False
        optimize_scale: True
        scale_lr: 0.001
        thresholds: [0.75, 0.5, 0.375, 0.25, 0.125, 0.09, 0.06, 0.04, 0.02, 0.005]
        weight_clip: True
        load_transform: True
        clip_version: v1
        reduce_memory: False
        scale_path: ../cache/activations/L32_3b/awq_w4g128
        clip_path: ../cache/activations/L32_3b/awq_w4g128
    quant_out: True
save:
    save_fp: False
    save_trans: False
    save_lightllm: False
    save_autogptq: False