tokenizer_config.json

{
  "clean_up_tokenization_spaces": true,  
  "auto_map": {
    "AutoTokenizer": ["sentencepiece_ja.SentencePieceJA"]
  }
}