config.json

{
  "model_type": "llama",
  "task_specific_params": {
    "text-generation": {
      "top_k": 40,
      "top_p": 0.6,
      "temperature": 0.2,
      "repetition_penalty": 1.1}
  }
}