Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Pytorch 2.0 #151

Closed
h-vetinari opened this issue Dec 5, 2022 · 15 comments
Closed

Pytorch 2.0 #151

h-vetinari opened this issue Dec 5, 2022 · 15 comments

Comments

@h-vetinari
Copy link
Member

Pytorch 2.0 recently got announced, and it looks like it'll bring lots of goodies.

However, with the strong focus on the new compiler backend, things are predictably getting heavier for the build. Seems the default backend will be triton, which we have in conda-forge already, but so far only for linux-x64 + CUDA (no CPU version). That feedstock is also pushing hard against the 6h limit.

Release is planned March 2023.

If we have the capacity, I think it'd make sense to start playing with this, perhaps even pushing into a dev branch, so that we might have a chance to provide feedback on critical issues before release.

@hmaarrfk
Copy link
Contributor

hmaarrfk commented Dec 6, 2022

We should likely create a branch for 1.13 and migrate it. While publishing rc to a dev channel. We can do this, even if just for the CPU builds.

@h-vetinari
Copy link
Member Author

IMO 1.13 can stay on main for a while, we only need a dev branch and pytorch_dev label. OTOH I think we could already start before the rc phase

@hmaarrfk
Copy link
Contributor

hmaarrfk commented Dec 8, 2022

ok. i won't branch out for a while then. Lets see how far somebody gets on the CIs. then if there is something for us to publish before the rc or official release, we can branch out then for 2.0.

I mostly like keeping the main branch at the latest and greatest version (according to upstream) and leaving branches for versions we deem important.

@h-vetinari
Copy link
Member Author

ok. i won't branch out for a while then

We should still IMO add a dev branch. We currently only have main and test - not sure what the latter was intended for

@hmaarrfk
Copy link
Contributor

hmaarrfk commented Dec 8, 2022

i think test was to test some CIs that isuru got that were associated with conda-forge.

@ngam
Copy link
Contributor

ngam commented Feb 17, 2023

Time to spin this forward? pytorch/pytorch#94937

@hmaarrfk
Copy link
Contributor

We could make a release candidate branch branch and release to a non main channel if you want

@ngam
Copy link
Contributor

ngam commented Feb 17, 2023

I meant a PR for testing to ensure we will be ~ready by the time it is released. Isn't the build supposed to be quite different now?

@h-vetinari
Copy link
Member Author

We can start with a PR; if we have the resources to create builds and push them to a dev channel, all the better

@h-vetinari
Copy link
Member Author

@Tobias-Fischer
Copy link
Contributor

Hi all,
Happy to help out if you need a hand. @hmaarrfk: You have quite a few draft/WIP PRs. Will it make sense to get some merged first, as this would simplify the version update? Or is it best to base a v2 branch on top of main?

@hmaarrfk
Copy link
Contributor

I've mostly given up on my experimental PRs. main is the best to follow.

@Tobias-Fischer
Copy link
Contributor

Just letting you know that I’ve been working on this and I’m pretty close, at least for the non-cuda Linux-64 build. Will create a PR next week

@h-vetinari
Copy link
Member Author

Closed by #165

@h-vetinari
Copy link
Member Author

Triton discussion continues in #166

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants