quantization_config.json

{
  "bits": 8,
  "data_type": "mx_fp",
  "group_size": 32,
  "sym": true,
  "enable_quanted_input": false,
  "autoround_version": "0.13.0",
  "block_name_to_quantize": "blocks,audio_injector.injector,audio_injector.injector_pre_norm_feat,audio_injector.injector_pre_norm_vec,audio_injector.injector_adain_layers",
  "quant_method": "auto-round",
  "packing_format": "auto_round:llm_compressor"
}