# Turn a classification model into seq2seq model?

**URL:** <https://discuss.huggingface.co/t/turn-a-classification-model-into-seq2seq-model/27487>\
**Category:** Beginners\
**Created:** [December 8, 2022, 8:11pm UTC](https://discuss.huggingface.co/t/turn-a-classification-model-into-seq2seq-model/27487 "2022-12-08T20:11:17Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![valentinwerner](https://avatars.discourse-cdn.com/v4/letter/v/ce73a5/32.png) [@valentinwerner](https://discuss.huggingface.co/u/valentinwerner)\
**Post date:** [December 8, 2022, 8:11pm UTC](https://discuss.huggingface.co/t/turn-a-classification-model-into-seq2seq-model/27487/1 "2022-12-08T20:11:17Z")

</div>

Hey everyone, I havent seen a similar question, so I am shooting my shot.

I would like to use transfer learning with a domain specific model. However, the most promising model does not support the AutoModelForSeq2Seq. Can I somehow still implement it as Seq2Seq model (e.g. by putting extra layers on top) or am I missing a critical architectural feature that would be needed to turn the model into an autoregressive seq2seq model?

I want to work with [snowood1/ConfliBERT-scr-uncased · Hugging Face](https://huggingface.co/snowood1/ConfliBERT-scr-uncased)

---

<div class="post-metadata">

**Author:** ![valentinwerner](https://avatars.discourse-cdn.com/v4/letter/v/ce73a5/32.png) [@valentinwerner](https://discuss.huggingface.co/u/valentinwerner)\
**Post date:** [December 11, 2022, 9:41am UTC](https://discuss.huggingface.co/t/turn-a-classification-model-into-seq2seq-model/27487/2 "2022-12-11T09:41:44Z")

</div>

Solved by using the EncoderDecoderModel.