quantization_config.json

{
  "bits": 4,
  "compute_dtype": "float16",
  "quantization_method": "bitsandbytes",
  "load_in_4bit": true,
  "load_in_8bit": false,
  "bnb_4bit_compute_dtype": "float16",
  "bnb_4bit_use_double_quant": true,
  "bnb_4bit_quant_type": "nf4"
}