-
Notifications
You must be signed in to change notification settings - Fork 7.1k
Upgrade nightly wheels to ROCm5.3 #6955
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
Upgrade nightly wheels to ROCm5.3 #6955
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The changes look good to me and the tests pass. Nevertheless I observed that some of the jobs such as Build Linux Wheels / pytorch/vision / manywheel-py3_10-rocm5_1_1 (pull_request)
still run for 5.1.1 which means there are more places that need to be modified. I think this is related to @osalpekar's work of migrating from CircleCI to GA.
I've added the RelEng team to confirm.
@datumbox Yes, I noticed that too. I filed a PR pytorch/test-infra#1219 for upgrading to ROCm5.3. cc @osalpekar |
@jithunnair-amd Thanks for confirming! @osalpekar Do you advise merging this PR or wait for the other fix? My understanding is that it can be merged but I could be wrong. Thanks! |
Thanks @jithunnair-amd and @datumbox. We should be good to merge both in any order! Note that the test-infra PR will be the one that actually changes the ROCm versions in the nightly wheels, since those are build via GHA. The remaining py3.7 CircleCI jobs that are changed by this PR are just for the docs build. |
@datumbox The test-infra PR is merged. Is this PR good to go now? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, thanks!
Hey @datumbox! You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py |
@datumbox @osalpekar While the PR pytorch/test-infra#1219 was merged on Dec 2, I see that rocm5.3 wheels haven't started being published yet for torchvision (https://download.pytorch.org/whl/nightly/rocm5.3/torchvision/). Even the latest rocm5.2 wheels for torchvision were only upto Nov 21 (https://download.pytorch.org/whl/nightly/rocm5.3/torchvision/). Would you know why? |
@atalman Any thoughts? |
@jithunnair-amd I see the Dec 5 rocm5.3 wheels at that page. They're closer to the top of the page for some reason (the ordering of the jobs on the page is a little strange). Here's the name and link for reference:
And successful build job: https://github.com/pytorch/vision/actions/runs/3620045031/jobs/6101821737. I see the Dec 5 rocm5.2 wheels on the nightly page as well (this time near the middle of the page): https://download.pytorch.org/whl/nightly/rocm5.2/torchvision/ |
@osalpekar perhaps I'm reading the package suffix name incorrectly, but this looks like a CPU build rather than ROCM accelerated one, isn't it? |
@malfet You're right, I see that issue and working on a fix now |
Just merged pytorch/test-infra#1230 in test-infra. Tests suggest that PR successfully fixes the issue @jithunnair-amd mentioned. I'll monitor the nightlies tomorrow to ensure they've been published correctly. |
Summary: * Update to ROCm 5.3 * Regenerate config.yml Reviewed By: datumbox Differential Revision: D41836896 fbshipit-source-id: f317897c7fc314698f8d42c00f09ff5d859cd9f0
Dependent on PR pytorch/pytorch#89101
cc @seemethere @jeffdaily