quantize_config.json

{
  "quantization": {
    "bits": 4,
    "group_size": 64,
    "only_transformer_blocks": true
  }
}