generation_config.json

{
  "bos_token_id": 128000,
  "do_sample": true,
  "eos_token_id": [
    128012
  ],
  "length_penalty": 1.0,
  "max_new_tokens": 128,
  "min_new_tokens": 0,
  "no_repeat_ngram_size": 0,
  "num_beams": 1,
  "pad_token_id": 128004,
  "repetition_penalty": 1.05,
  "temperature": 0.7,
  "transformers_version": "5.0.0.dev0",
  "use_cache": true
}