transfer learning and answer prediction on fill-in-the-blank type of QA dataset

[tayalps]

fill-in-the-blank questions could be treated as questions with a language model where the fill in the blank is like applying the mask to the word.
How could I use haystack for that? I have software domain-related QA data, so how could i apply transfer learning and predict the answers correctly?

[tholor]

Hey @tayalps ,
fill-in-the-blank questions are currently not supported in haystack. Can you elaborate a bit on your use case? So far we haven't seen many applications for this QA style in the industry, but maybe we are missing something here.
In any case, it should be fairly simple to take a model from Transformers or FARM and just predict the blanks. Happy to support you along the way in case further questions pop up.

[tholor]

Closing this as "fill-in-the-blank questions" are currently not in scope for haystack. @tayalps feel free to reopen and clarify the use case here.