quantize_config.json

{
  "bits": 4,
  "group_size": 64,
  "sym": false,
  "desc_act": false,
  "quant_method": "gptq",
  "true_sequential": true,
  "model_name_or_path": "deepseek-ai/deepseek-coder-6.7b-instruct",
  "model_file_base_name": "model",
  "checkpoint_format": "gptq_v2",
  "meta": {
    "recommended_backend": "exllama_v2",
    "marlin_compatible": false
  }
}