generation_config.json

{
    "chat_format": "chatml",
    "eos_token_id": 0,
    "pad_token_id": 0,
    "max_window_size": 4096,
    "max_new_tokens": 4096,
    "do_sample": true,
    "top_k": 0,
    "top_p": 0.1,
    "repetition_penalty": 1.0,
    "transformers_version": "4.31.1"
}