Remove usage of patched tokenizer for training

tokenizer_config.json

@@ -19,6 +19,6 @@
   "special_tokens_map_file": null,
   "src_lang": "jpn_Jpan",
   "tgt_lang": "eng_Latn",
-  "tokenizer_class": "TrainingNllbTokenizer",
+  "tokenizer_class": "NllbTokenizer",
   "unk_token": "<unk>"
 }