{
  "quantization_stats": {
    "original_size_mb": 6362.53515625,
    "quantized_size_mb": 2223.747215270996,
    "size_reduction_mb": 4138.787940979004,
    "size_reduction_percent": 65.04935280260761,
    "compression_ratio": 2.8611773463085064,
    "n_quantized_layers": 2342
  },
  "safetensors_size_mb": 3352.126941680908,
  "mbq_config": {
    "w_bit": 4,
    "a_bit": 8,
    "mixed_precision": true,
    "sensitivity_metric": "hessian",
    "calibration_samples": 128,
    "preserve_ratio": 0.15
  },
  "model_info": {
    "base_model": "deepseek-ai/DeepSeek-OCR",
    "total_params": 3336106240,
    "quantized_layers": 2342
  },
  "bit_allocation_summary": {
    "8-bit": 351,
    "4-bit": 1991
  },
  "files": {
    "model.safetensors": "3352.13 MB",
    "quantized_weights.pt": "Compressed quantized weights",
    "config.json": "Model configuration",
    "tokenizer files": "Tokenizer configuration"
  }
}