# About Transformer task prefix

**URL:** <https://discuss.huggingface.co/t/about-transformer-task-prefix/5985>\
**Category:** Beginners\
**Created:** [May 4, 2021, 9:15pm UTC](https://discuss.huggingface.co/t/about-transformer-task-prefix/5985 "2021-05-04T21:15:12Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Cheligeer](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/cheligeer/32/3475_2.png) [@Cheligeer](https://discuss.huggingface.co/u/Cheligeer)\
**Post date:** [May 4, 2021, 9:15pm UTC](https://discuss.huggingface.co/t/about-transformer-task-prefix/5985/1 "2021-05-04T21:15:12Z")

</div>

Hi all,

I am a new user, please forgive me if my question is too silly, but I tried many searches and still cannot find a satisfactory solution so here I am.

I want to fine-tune the T5 model with my sequence to sequence task that take one sentence as input and generate another sentence correspondingly. I read many examples from huggingface community notebooks and other sources, and i found there is a prefix added to the input or target.  
prefix at input [article](https://towardsdatascience.com/asking-the-right-questions-training-a-t5-transformer-model-on-a-new-task-691ebba2d72c).  
prefix at output [notebook](https://github.com/abhimishra91/transformers-tutorials/blob/master/transformers_summarization_wandb.ipynb).

Can I build whatever prefix by myself ? e.g. ‘paraphrase\_sentence’, or it is a something already fixed by authors ?  
I also want to know where should I add the prefix for my case (input or output) ? and what prefix should I add for my case?

my data is like:  
sentence\_a1, sentence\_b1,  
sentence\_a2, sentence\_b2,  
…

and I would to train a model that can generate a sentence to a given sentence.  
something similar to paraphrase.