Open aatkinson opened 5 years ago
Great questions :)
As for the road map, we are mainly working on 5 main areas (roughly sorted according to priority):
For the release cadence, we will do our best to have a release every month. And yeah as we mentioned in our release blog post here this project powers some of the core projects in production at Facebook and it indeed has a substantial backing from different teams in Facebook AI.
Neat, this is all good to know, thanks Ahmed.
I'm looking forward to seeing more model and task examples, particularly advanced ones.
Also that release blog post link appears to be broken for me, or at least inaccessible. I'm assuming it's the same content as this one? https://code.fb.com/ai-research/pytext-open-source-nlp-framework/.
Yes, this is the right link.
Thanks for the answer @ahhegazy! Does integration with fairseq fall under the 1st area?
Thanks for the answer @ahhegazy! Does integration with fairseq fall under the 1st area?
I am curious about the fairest integration as well. Any plans on that?
Yes, we are working on integrating with translate: https://github.com/pytorch/translate/tree/master/pytorch_translate which is Fairseq models with production support
Hello @ahhegazy, and thanks for sharing this amazing work, just a quick question, can we have an idea about timing regarding fairseq integration ? Thanks a lot !
@ahhegazy Another quick question, does multilingual modeling support is still on the radar? Also really interested regarding fairseq timing question.
And thanks for sharing this codebase with the community!
Hi,
The goals are ambitious, the codebase is in flux, and future directions are outlined in your paper https://arxiv.org/pdf/1812.08729.pdf, but it's unclear what the roadmap is for this project.
What release cadence can we expect, and what features are prioritized?
I'm also curious if this has a substantial backing by Facebook / the PyTorch team.
Thanks