generation_config.json

{
    "chat_format": "chatml",
    "max_window_size": 24000,
    "max_new_tokens": 2000,
    "do_sample": true,
    "top_k": 0,
    "top_p": 0.8,
    "repetition_penalty": 1.1,
    "transformers_version": "4.31.0"
  }