# Does the tokenization in BERT change after fine-tuning?

**URL:** <https://discuss.huggingface.co/t/does-the-tokenization-in-bert-change-after-fine-tuning/30580>\
**Category:** Models\
**Created:** [January 27, 2023, 6:12pm UTC](https://discuss.huggingface.co/t/does-the-tokenization-in-bert-change-after-fine-tuning/30580 "2023-01-27T18:12:45Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![siddharthtumre](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/siddharthtumre/32/5865_2.png) [@siddharthtumre](https://discuss.huggingface.co/u/siddharthtumre)\
**Post date:** [January 27, 2023, 6:12pm UTC](https://discuss.huggingface.co/t/does-the-tokenization-in-bert-change-after-fine-tuning/30580/1 "2023-01-27T18:12:45Z")

</div>

I am using a BERT based model. I have fine-tuned on a dataset for few epochs and saved the trained model.  
I am loading my tokenizer and my model from the saved checkpoint and I could see that the tokenization happening at this checkpoint is different from the one that is not fine-tuned.

I am not sure if this is the expected behaviour.

Thanks.
