So I was wondering if I could use the language model as a base to build a question-answering model using the SQUad dataset. I’m not quite sure as to how to approach this problem. Any suggestions on what could be a possible approach would be much appreciated.
You may want to consider using a seq2seq or attentional model instead of ULMFiT. See Lesson 11 (I believe) for an example of how to setup the former.
I still don’t understand the differences between ELMo and ULMFiT, so this may be irrelevant. However, in my mind they are similar, and I know ELMo does better than previous SOTA on SQuAD: https://github.com/allenai/document-qa/tree/master/docqa/elmo
So it seems like a reasonable idea to me… I’d check that repo for how they do it with ELMo
Hi, wondering if you had any success with this? I’m assuming that you’d use a text classifier to perform QA. How did you layout the data frame?
Hey, any progress or ideas on this?