generation_config.json

{
  "_from_model_config": true,
  "bos_token_id": 128000,
  "eos_token_id": 128001,
  "max_new_tokens": 256,
  "max_time": 120.0,
  "min_new_tokens": 2,
  "pad_token_id": 128001,
  "transformers_version": "4.38.2"
}