{
  "quantization_method": "mixed_precision_nf4",
  "model_type": "sd",
  "description": "First and last blocks kept at bfloat16, middle layers quantized to NF4 (sd architecture)",
  "high_precision_layers_count": 99
}