{
  "model_max_length": 128,
  "special_tokens": [
    "<pad>",
    "[UNK]",
    "<s>",
    "</s>",
    "<cls>",
    "<sep>",
    "<mask>"
  ],
  "tokenizer_class": "PreTrainedTokenizerFast"
}