quantization_config.json

{
  "bits": 4,
  "group_size": 128,
  "sym": false,
  "data_type": "int",
  "seqlen": 512,
  "batch_size": 2,
  "gradient_accumulate_steps": 4,
  "autoround_version": "0.6.0",
  "quant_method": "awq",
  "zero_point": true,
  "version": "gemm",
  "modules_to_not_convert": [
    "lm_head"
  ]
}