Finetune a model Q/A Response

[Btibert3]

It's not obvious to me if this sort of task would be supported here, but wanted to ask in case I was missing something in the flow in the library.

In short, I have a domain-specific dataset that is a single-turn Question and Answer. Think a webinar and a viewer asks a question to the presenter, who can send back a single response Answer to the Question. I was thinking about how I might be able to retrain/tine-tune a model (say GPT2) to help learn how to generate responses to the common questions. I was framing this as a machine translation task, though the language of the inputs passed into the encoder and decoder sections would be the same, just Question -> encoder, known Answer to -> decoder.

Is this task possible using finetune?