quantize_config.json

{
  "bits": 4,
  "group_size": 128,
  "damp_percent": 0.1,
  "desc_act": false,
  "static_groups": false,
  "sym": true,
  "true_sequential": false,
  "model_name_or_path": "/home/alexm/models/marlin_24/tinyllama-24-int-src",
  "model_file_base_name": "model",
  "quant_method": "gptq",
  "checkpoint_format": "marlin_24"
}