transformer/quantization_config.json

{
  "bits": 4,
  "data_type": "int",
  "group_size": 128,
  "sym": true,
  "batch_size": 1,
  "gradient_accumulate_steps": 8,
  "low_gpu_mem_usage": true,
  "autoround_version": "0.13.0",
  "block_name_to_quantize": "noise_refiner,context_refiner,layers",
  "quant_method": "auto-round",
  "packing_format": "auto_round:auto_gptq"
}