generation_config.json

{
  "attn_implementation": "flash_attention_2",
  "bos_token_id": 128000,
  "do_sample": true,
  "eos_token_id": 128001,
  "max_length": 4096,
  "num_assistant_tokens": 5,
  "num_assistant_tokens_schedule": "heuristic",
  "return_legacy_cache": true,
  "temperature": 0.6,
  "top_p": 0.9,
  "transformers_version": "4.45.1"
}