How to make RoBERTa question answer models take tokens from the question instead of the context?

I would like to do this to use multiple choice question answer datasets