# BertForSequenceClassification classification head question

**URL:** <https://discuss.huggingface.co/t/bertforsequenceclassification-classification-head-question/20104>\
**Category:** 🤗Transformers\
**Created:** [July 7, 2022, 11:50am UTC](https://discuss.huggingface.co/t/bertforsequenceclassification-classification-head-question/20104 "2022-07-07T11:50:26Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![bilionz](https://avatars.discourse-cdn.com/v4/letter/b/ee7513/32.png) [@bilionz](https://discuss.huggingface.co/u/bilionz)\
**Post date:** [July 7, 2022, 11:50am UTC](https://discuss.huggingface.co/t/bertforsequenceclassification-classification-head-question/20104/1 "2022-07-07T11:50:26Z")

</div>

Hello there,

In [BERT](https://huggingface.co/docs/transformers/model_doc/bert#transformers.BertForSequenceClassification)  
it is mentioned that the classifiaction head is “(a linear layer on top of the pooled output)”.

I understand that the pooled output is among all the outputs and not just the CLS token (or else it wouldnt be pooled I guess).

I would like some confirmation on this because it is not 100% clear for readers.

Thanks in advance.