# Train T5 decoder only on a different language

**URL:** <https://discuss.huggingface.co/t/train-t5-decoder-only-on-a-different-language/4435>\
**Category:** Models\
**Created:** [March 16, 2021, 6:12am UTC](https://discuss.huggingface.co/t/train-t5-decoder-only-on-a-different-language/4435 "2021-03-16T06:12:42Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Masum](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/masum/32/1230_2.png) [@Masum](https://discuss.huggingface.co/u/Masum)\
**Post date:** [March 16, 2021, 6:12am UTC](https://discuss.huggingface.co/t/train-t5-decoder-only-on-a-different-language/4435/1 "2021-03-16T06:12:42Z")

</div>

I would like to train the T5 decoder in a different language with a different tokenization scheme and take advantage of the pretrained encoder’s ability to encode English. Is it possible to do with the Huggingface library? Any pointers?

Thanks in advance!