# Impossible questions when finetuning QA models

**URL:** <https://discuss.huggingface.co/t/impossible-questions-when-finetuning-qa-models/12068>\
**Category:** 🤗Transformers\
**Created:** [November 19, 2021, 5:49pm UTC](https://discuss.huggingface.co/t/impossible-questions-when-finetuning-qa-models/12068 "2021-11-19T17:49:21Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![erickrf](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/erickrf/32/239_2.png) [@erickrf](https://discuss.huggingface.co/u/erickrf)\
**Post date:** [November 19, 2021, 5:49pm UTC](https://discuss.huggingface.co/t/impossible-questions-when-finetuning-qa-models/12068/1 "2021-11-19T17:49:21Z")

</div>

How can I encode impossible questions (i.e., without an answer in the context) when finetuning a QA model? I got a little confused with the code, not sure if giving start and end position as 0 (pointing to the `[CLS]` token) is the right approach.