Use Japanese tokenizer with proper config

tokenizer_config.json

@@ -6170,6 +6170,7 @@
     }
   },
   "clean_up_tokenization_spaces": false,
+  "extra_special_tokens": {},
   "model_max_length": 1000000000000000019884624838656,
   "tokenizer_class": "PreTrainedTokenizerFast"
 }