# Can I train ELECTRA from scratch using hugging face?

**URL:** <https://discuss.huggingface.co/t/can-i-train-electra-from-scratch-using-hugging-face/71371>\
**Category:** Models\
**Created:** [January 31, 2024, 5:37pm UTC](https://discuss.huggingface.co/t/can-i-train-electra-from-scratch-using-hugging-face/71371 "2024-01-31T17:37:52Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![gamster2](https://avatars.discourse-cdn.com/v4/letter/g/bcef8e/32.png) [@gamster2](https://discuss.huggingface.co/u/gamster2)\
**Post date:** [January 31, 2024, 5:37pm UTC](https://discuss.huggingface.co/t/can-i-train-electra-from-scratch-using-hugging-face/71371/1 "2024-01-31T17:37:52Z")

</div>

I can train BERT from scratch using PyTorch code from hugging face GitHub.

However, when it comes to ELECTRA, I just cannot find PyTorch resources.

How can I use get labels of 1 or 0 indicating whether token is replaced or not?

Please somebody help me…
