# Deberta v3 Input length and Absolute positional embeddings

**URL:** <https://discuss.huggingface.co/t/deberta-v3-input-length-and-absolute-positional-embeddings/56960>\
**Category:** Models\
**Created:** [September 30, 2023, 8:09pm UTC](https://discuss.huggingface.co/t/deberta-v3-input-length-and-absolute-positional-embeddings/56960 "2023-09-30T20:09:15Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![venetis](https://avatars.discourse-cdn.com/v4/letter/v/90ced4/32.png) [@venetis](https://discuss.huggingface.co/u/venetis)\
**Post date:** [September 30, 2023, 8:09pm UTC](https://discuss.huggingface.co/t/deberta-v3-input-length-and-absolute-positional-embeddings/56960/1 "2023-09-30T20:09:15Z")

</div>

Hello, I have two questions regarding the implementation of deberta v3.

Given that the masked language modelling has been replaced with an Electra style learning, is it safe to assume that the absolute positonal embeddings have been dropped?

Furthermore, if that is the case, is it safe to assume the V3 version can deal with any input length?