-
Notifications
You must be signed in to change notification settings - Fork 7.2k
Add CircleCI (v2) #1298
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
Add CircleCI (v2) #1298
Conversation
@soumith @ezyang I have the impression that I can entirely remove https://github.com/pytorch/vision/tree/master/packaging/conda , plus the build instructions in https://github.com/pytorch/vision/blob/master/packaging/README.md seem to be deprecated. @ezyang does this sounds about right? (for Linux and OSX at least) |
It's OK for Linux and OS X; but it was previously used by Windows, unless peterjc123 has fixed the Windows pipeline. |
CUDA tests via conda build is a useful thing to test for. However, I'm not sure I would have them be the only testing we do (since builds always happening inside |
@peterjc123 This has windows bits now. (My audio PR isn't that useful, I never got it to actually work haha) |
Note that the Windows bits are not doing anything useful yet... but I'd love to be able to run some (at least CPU) tests on Windows, but I have no Windows experience |
@peterjc123 awesome, thanks a lot for all the help! Let me know once you are done with the Windows part, I'll move your changes to |
Also, I'll disable the tests for Windows that are failing, don't worry about it |
@peterjc123 I have a commit ready to be pushed that disables the failing tests for Windows for now. Let me know when I can push it, to avoid potential conflicts |
@fmassa Sure, feel free to push that. Current build failure is caused by something else. (Conda base package not uploaded but cpu-only packages uploaded) |
@peterjc123 yes, that's true, and it is the cause of the failing builds that we currently have. |
@fmassa Now it won't fail when only part of new packages are uploaded because I introduced fine-grained search for latest PyTorch version. |
Re-do of #1295, but re-using most of the packaging code that we already have. Takes into account the comments of @ezyang from #1295
The final code is much simpler as well it seems.
Tests currently are forced to run on the CPU because we currently do not have PyTorch nightly conda packages for Linux with CUDA
I still need to improve the job selection to run in master / CI