# Finetune GPT2 in tensorflow on custom data example programmatically

**URL:** <https://discuss.huggingface.co/t/finetune-gpt2-in-tensorflow-on-custom-data-example-programmatically/439>\
**Category:** Beginners\
**Created:** [July 23, 2020, 7:46am UTC](https://discuss.huggingface.co/t/finetune-gpt2-in-tensorflow-on-custom-data-example-programmatically/439 "2020-07-23T07:46:41Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![georgedittmar](https://avatars.discourse-cdn.com/v4/letter/g/ecccb3/32.png) [@georgedittmar](https://discuss.huggingface.co/u/georgedittmar)\
**Post date:** [July 23, 2020, 7:46am UTC](https://discuss.huggingface.co/t/finetune-gpt2-in-tensorflow-on-custom-data-example-programmatically/439/1 "2020-07-23T07:46:41Z")

</div>

I am trying to utilize GPT2 for text generation in tf 2 and test out some basic fine tuning on custom data but the examples I have found only seem to utilize the scripts to do this, or assume only wanting to fine tune a classification task. Are there any examples floating around of trying to do fine tuning of the model over custom text data to create new text generation output in tf2? I am not sure if I just need to set it up like any other sequence model like I would an lstm or if GPT2 should just take in the tokenized data after using the gpt2 tokenizer?