{
  "quantization": {
    "group_size": 64,
    "bits": 4
  },
  "precision": "int4"
}