quantize_config.json

{
  "bits": 4,
  "desc_act": true,
  "group_size": 128,
  "model_file_base_name": "model"
}