search

artitw / BERT_QA

Accelerating the development of question-answering systems based on BERT and TF 2.0
https://arxiv.org/abs/1909.05017
Apache License 2.0
19 stars 4 forks source link

Question Generation #1

Open DivyaWadehra opened 4 years ago

DivyaWadehra commented 4 years ago

Does this solve question generation problem?

artitw commented 4 years ago

This is for evaluating the questions generated using SQuAD. However, if there is interest in the question generation part, let me know and I'll see what I can do.

DivyaWadehra commented 4 years ago

Yes, I want the code for question generation

felicitywang commented 4 years ago

@artitw Thanks for the great work. Is there any easy way to do inferencing and play with it(e.g. inferencing code + trained model or via an online demo etc.)?

artitw commented 4 years ago

@felicitywang @DivyaWadehra appreciate your interest in question generation very much. I will look into open-sourcing it as soon as I can. Currently it's not in an easily consumable form.

DivyaWadehra commented 4 years ago

Thankyou Looking forward for the code

artitw commented 4 years ago

@DivyaWadehra @felicitywang check out https://github.com/artitw/text2text

felicitywang commented 4 years ago

@artitw Thank you. However, this project is a pure wrapper of Microsoft's UniLM, which you didn't cite in your paper - is this UniLM the model you used in your paper?

artitw commented 4 years ago

Yes, that repo is based on UniLM. I think they released a good pretrained model we can easily play with. Ours is based on vanilla transformer; see https://github.com/artitw/tensor2tensor Currently I don't have the bandwidth to build an online demo for it. :(