# Learning sets and disabling positional embedding knowledge?

**URL:** <https://discuss.huggingface.co/t/learning-sets-and-disabling-positional-embedding-knowledge/39434>\
**Category:** Intermediate\
**Created:** [May 10, 2023, 6:57pm UTC](https://discuss.huggingface.co/t/learning-sets-and-disabling-positional-embedding-knowledge/39434 "2023-05-10T18:57:39Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![BadDepartment](https://avatars.discourse-cdn.com/v4/letter/b/c57346/32.png) [@BadDepartment](https://discuss.huggingface.co/u/BadDepartment)\
**Post date:** [May 10, 2023, 6:57pm UTC](https://discuss.huggingface.co/t/learning-sets-and-disabling-positional-embedding-knowledge/39434/1 "2023-05-10T18:57:39Z")

</div>

I have an NLP task that I would like to use a transformer against but the position of my tokens is arbitrary; is there any way to disable positional learning during fine-tuning? if not, should I feature engineer my text (basically a set) such that it has some sort of positional relationship?