generation_config.json

{
  "max_length": 4096,
  "temperature": 0.8,
  "top_p": 0.95,
  "top_k": 40,
  "repetition_penalty": 1.1,
  "do_sample": true,
  "num_return_sequences": 1,
  "early_stopping": true,
  "length_penalty": 1.0,
  "no_repeat_ngram_size": 2,
  "presence_penalty": 0.3,
  "frequency_penalty": 0.3,
  "min_length": 10,
  "use_cache": true
}