q8/mllm/quantize_config.json

{
  "bits": 8,
  "quantization": { "group_size": 64 }
}