quantize_config.json

{
  "bits": 4,
  "group_size": 128,
  "damp_percent": 0.1,
  "desc_act": true,
  "static_groups": false,
  "sym": true,
  "true_sequential": true,
  "model_name_or_path": "/fsx/llama405b/4bit-gptq-instruct",
  "model_file_base_name": "gptq_model-4bit-128g",
  "is_marlin_format": false,
  "quant_method": "gptq"
}