tokenizer_config.json

{
  "clean_up_tokenization_spaces": true,
  "model_max_length": 512,
  "special_tokens": [
    "<s>",
    "<pad>",
    "</s>",
    "<unk>",
    "<cls>",
    "<sep>",
    "<mask>"
  ],
  "tokenizer_class": "PreTrainedTokenizerFast"
}