generation_config.json

{
  "chat_format": "chatml",
  "do_sample": true,
  "eos_token_id": 0,
  "max_new_tokens": 4096,
  "max_window_size": 4096,
  "pad_token_id": 0,
  "top_k": 0,
  "top_p": 0.1,
  "transformers_version": "4.38.2"
}