# What's the deal? Why do I need another tokenizer config for MT5?
As some of you might have already noticed, currently the official release of the tokenizer sentencepiece model of the MT5 family is misconfigured, resulting in finetuning errors and strange generations when fine-tuning for the span corruption task. This has meant at least in my part a non-trivial amount of debugging the model until finally finding what's wrong, as previously mentioned in a [forum question](https://huggingface.co/google/mt5-small/discussions/8), the issue lies in the tokenizer not properly tokenizing the sentinel tokens. Resulting in the breaking of the sentinel tokens into smaller tokens when they should have been kept intact. Long story short, the sentinel tokens were saved in the sentencepiece model with an empty space prefix (e.g., " <extra_id_N>" instead of "<extra_id_N>") resulting in all this mess, this meant that I had to manually change the sentencepiece model sentinel tokens into the proper for…
