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

ROCM binaries aren't built with jpeg support #8079

Closed
NicolasHug opened this issue Oct 30, 2023 · 7 comments
Closed

ROCM binaries aren't built with jpeg support #8079

NicolasHug opened this issue Oct 30, 2023 · 7 comments

Comments

@NicolasHug
Copy link
Member

NicolasHug commented Oct 30, 2023

The jobs started failing on this line:

torch.ops.image._jpeg_version() = -1
  File "pytorch/vision/test/smoke_test.py", line 86, in main
    assert torch.ops.image._is_compiled_against_turbo()
AssertionError

but the error originates from this line

Building torchvision without JPEG image support

This is happening on both main and on the release branch, so we need to fix this before the bugfix release.

CC @atalman

cc @jeffdaily @jithunnair-amd

@jeffdaily
Copy link
Contributor

#8080 did not fix this. What are our next steps?

@jeffdaily
Copy link
Contributor

Do we need to regenerate the pytorch/manylinux-builder:rocm5.6 image? Is it a problem in this image?

@NicolasHug
Copy link
Member Author

@jeffdaily honestly I'm not sure. Nothing has changed on the torchvision side so the problem likely comes from somewhere else. I have asked our release engineering team to take a look.

@malfet
Copy link
Contributor

malfet commented Oct 30, 2023

I guess problem is longstanding, but docker update actually cased an issue. Confirmed now:
This used to pass https://github.com/pytorch/vision/actions/runs/6666267146/job/18117478178 but after a re-run it fails with the same signature: https://github.com/pytorch/vision/actions/runs/6666267146/job/18194624308
So I suspect it probably side-effect of pytorch/builder#1572
@jeffdaily speaking of, do you run any tests on your end after final release candidate has been build?

@jeffdaily
Copy link
Contributor

@malfet no, we don't run any tests on our end. We rely on upstream CI infra in this case.

@NicolasHug
Copy link
Member Author

OK, if this is a docker issue then I think we'll have to disable those tests on torchvision until this gets resolved. I'll keep this issue open, LMK if there is any progress on this!

@NicolasHug
Copy link
Member Author

OK, looks like the latest Rocm builds + smoke tests are fine on main and the release branch now.

Probably fixed by pytorch/builder#1575 ? Thanks a ton @huydhn!

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

No branches or pull requests

3 participants