{
  "tokenizer_class": "LlamaTokenizer",
  "clean_up_tokenization_spaces": false
}