# Custom PostProcessor?

**URL:** <https://discuss.huggingface.co/t/custom-postprocessor/25847>\
**Category:** 🤗Tokenizers\
**Created:** [November 10, 2022, 6:46am UTC](https://discuss.huggingface.co/t/custom-postprocessor/25847 "2022-11-10T06:46:40Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![tonyswoo](https://avatars.discourse-cdn.com/v4/letter/t/e495f1/32.png) [@tonyswoo](https://discuss.huggingface.co/u/tonyswoo)\
**Post date:** [November 10, 2022, 6:46am UTC](https://discuss.huggingface.co/t/custom-postprocessor/25847/1 "2022-11-10T06:46:40Z")

</div>

Is there a way to implement a custom post-processor?

On this forum, I did see people discussing [how to implement a custom pre-tokenizer and normalizer](https://discuss.huggingface.co/t/1637), but I haven’t seen any posts discussing the post-processor.

My first guess was to subclass `PostProcessor` and implement `num_special_tokens_to_add` and `process`, but trying to initialize the subclass just gives a `TypeError: No constructor defined`.  
(Assuming it has something to do with the Rust-Python binding and not having a ` __new__ ` method)

So, does the HuggingFace tokenizers library support custom post-processing? If so, is there any instruction or example code available?

Thank you.