checkpoint-20000/tokenizer_config.json

{
  "clean_up_tokenization_spaces": true,
  "model_max_length": 512,
  "tokenizer_class": "MarindaTokenizer"
}