# Remove columns before training

**URL:** <https://discuss.huggingface.co/t/remove-columns-before-training/28908>\
**Category:** Beginners\
**Created:** [January 4, 2023, 7:36am UTC](https://discuss.huggingface.co/t/remove-columns-before-training/28908 "2023-01-04T07:36:50Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![cravephi](https://avatars.discourse-cdn.com/v4/letter/c/d78d45/32.png) [@cravephi](https://discuss.huggingface.co/u/cravephi)\
**Post date:** [January 4, 2023, 7:36am UTC](https://discuss.huggingface.co/t/remove-columns-before-training/28908/1 "2023-01-04T07:36:50Z")

</div>

Hi,

I see in many examples that we remove unnecessary columns from the dataset, before training.  
Is it mandatory? I guess Trainer will take the columns produced by the tokenizer anyway (together with the “labels” columns).  
Is it to save some IOs? I guess transformer does not load unnecessary columns, but not sure then, why in examples, we remove them.

Sorry, very beginner question here 🙂