quantization_config.json

{
  "quant_method": "awq",
  "modules_to_not_convert": null,
  "version": "gemm",
  "iters": 200,
  "lr": 0.005,
  "minmax_lr": 0.005,
  "enable_minmax_tuning": true,
  "enable_quanted_input": true,
  "scale_dtype": "float16",
  "sym": false,
  "bits": 4,
  "group_size": 128,
  "zero_point": true
}