generation_config.json

{
  "_from_model_config": true,
  "bos_token_id": 0,
  "do_sample": true,
  "eos_token_id": 50256,
  "max_new_tokens": 20,
  "output_attentions": false,
  "output_hidden_states": false,
  "pad_token_id": 50256,
  "repetition_penalty": 1.0,
  "stop_strings": [
    "<|endoftext|>",
    "<|im_end|>",
    "</s>"
  ],
  "temperature": 0.8,
  "top_k": 50,
  "top_p": 0.95,
  "transformers_version": "5.8.1",
  "use_cache": true
}