quantize_config.json

{
  "bits": 4,
  "group_size": 128,
  "sym": false,
  "data_type": "int",
  "enable_quanted_input": true,
  "enable_minmax_tuning": true,
  "seqlen": 512,
  "batch_size": 4,
  "scale_dtype": "torch.float16",
  "lr": 0.005,
  "minmax_lr": 0.005,
  "gradient_accumulate_steps": 1,
  "iters": 200,
  "amp": false,
  "nsamples": 128,
  "low_gpu_mem_usage": false,
  "to_quant_block_names": [
    [
      "gpt_neox.layers.0",
      "gpt_neox.layers.1",
      "gpt_neox.layers.2",
      "gpt_neox.layers.3",
      "gpt_neox.layers.4",
      "gpt_neox.layers.5",
      "gpt_neox.layers.6",
      "gpt_neox.layers.7",
      "gpt_neox.layers.8",
      "gpt_neox.layers.9",
      "gpt_neox.layers.10",
      "gpt_neox.layers.11",
      "gpt_neox.layers.12",
      "gpt_neox.layers.13",
      "gpt_neox.layers.14",
      "gpt_neox.layers.15",
      "gpt_neox.layers.16",
      "gpt_neox.layers.17",
      "gpt_neox.layers.18",
      "gpt_neox.layers.19",
      "gpt_neox.layers.20",
      "gpt_neox.layers.21",
      "gpt_neox.layers.22",
      "gpt_neox.layers.23",
      "gpt_neox.layers.24",
      "gpt_neox.layers.25",
      "gpt_neox.layers.26",
      "gpt_neox.layers.27",
      "gpt_neox.layers.28",
      "gpt_neox.layers.29",
      "gpt_neox.layers.30",
      "gpt_neox.layers.31",
      "gpt_neox.layers.32",
      "gpt_neox.layers.33",
      "gpt_neox.layers.34",
      "gpt_neox.layers.35"
    ]
  ],
  "enable_norm_bias_tuning": false,
  "autoround_version": "0.4.1",
  "quant_method": "gptq",
  "desc_act": false,
  "true_sequential": false,
  "damp_percent": 0.01
}