Support for HF's transformers 3.1+

[tingofurro]

Thanks for an excellent library.
Any plan to support the new model APIs that happened in transformers==3.1.0 and beyond?
Would it require major work or are there pointers on how to adapt it?

Thank you!

[feihugis]

@tingofurro Thanks for your interest. We did not get a chance to check the new APIs of transformers (>= 3.1.0) yet. But I think most of the works would be to fix the broken API changes in the new versions of transformers, and there should be no big change in the core optimization code. If you are interested in this feature, please feel free to submit a PR to support it. I'm glad to provide support if needed.

[feihugis]

@tingofurro FYI, we are working on supporting the latest version of transformers. May I ask which models of transformers you are interested in?

[tingofurro]

Hey @feihugis that's great news I would still very much benefit from it!
I mostly use GPT2 and BART and I think these are some of the most popular.

[feihugis]

Thanks @tingofurro! We are working on GPT2 and BART. Will keep you updated.

[JulianneKnott]

@tingofurro Fastseq supports transformers versions 4.12.0 - 4.12.5 now. Let us know if you have any feedback!

[tingofurro]

That's great, I'll test it out in the next few days, it should be very helpful!