# Code example of getting cross attention from T5?

**URL:** <https://discuss.huggingface.co/t/code-example-of-getting-cross-attention-from-t5/31740>\
**Category:** Intermediate\
**Created:** [February 15, 2023, 5:16am UTC](https://discuss.huggingface.co/t/code-example-of-getting-cross-attention-from-t5/31740 "2023-02-15T05:16:21Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![veritas2019](https://avatars.discourse-cdn.com/v4/letter/v/e47774/32.png) [@veritas2019](https://discuss.huggingface.co/u/veritas2019)\
**Post date:** [February 15, 2023, 5:16am UTC](https://discuss.huggingface.co/t/code-example-of-getting-cross-attention-from-t5/31740/1 "2023-02-15T05:16:21Z")

</div>

Hi fellow hackers – could you point me to a short code snippet showing how to obtain cross attention weights from a T5 model (T5ForConditionalGeneration)? I’d like to get the attention distribution for every decoded token on all input tokens.

Thanks a lot!