generation_config.json

{
  "alpha": 0.0,
  "attn_implementation": "flash_attention_2",
  "beta": 1.0,
  "bos_token_id": 128000,
  "do_sample": true,
  "eos_token_id": [
    128001,
    128008,
    128009
  ],
  "gamma": 1.0,
  "kl_temperature": 1.0,
  "max_length": 131072,
  "pad_token_id": 128001,
  "temperature": 0.4,
  "top_p": 0.9,
  "transformers_version": "4.46.3"
}