wheels CI: update to torch 2.10 for 'oldest' configuration#5485
wheels CI: update to torch 2.10 for 'oldest' configuration#5485jameslamb merged 1 commit intorapidsai:mainfrom
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
|
/ok to test |
| with: | ||
| build_type: pull-request | ||
| script: ci/test_wheel_cugraph.sh | ||
| wheel-tests-cugraph-nightly: |
There was a problem hiding this comment.
Just doing this temporarily to test that this will actually fit the nightlies.
If this works and the PR's approved, I'll push a commit reverting this out, then admin-merge this to avoid needing to wait on another CI cycle.
|
Based on offline discussion... we'll re-target this to re-targeted to |
f39f28c to
97ea8a6
Compare
|
Admin-merging this, the test failures are expected to be resolved in #5487 |
|
@jameslamb this should be fixed now. Please see: pytorch/pytorch#179821 (comment) |
|
Thank you! I'll get this reverted. |
…apidsai#5485)" This reverts commit 6ad9663.
…5485) (#5493) Reverts #5485 We can now install `torch` 2.9 CUDA 12.9 wheels again, the issue on the PyTorch `pip` index has been resolved: pytorch/pytorch#179821 (comment)
Nightly wheel tests on the "oldest" configuration started failing yesterday, like this:
(build link)
This looks like an issue with the metadata on
torch's index: pytorch/pytorch#145501 (comment)For now, this proposes just testing against
torch-2.10.0for the "oldest" configuration, to get CI working again.Notes for Reviewers
How I tested this
On an earlier commit, ran the full nightly matrix on
release/26.04. Saw it pass: https://github.com/rapidsai/cugraph/actions/runs/24143566958/job/70480989274This change isn't RAPIDS-branch-specific, so I'm confident this will resolve this issue in the nightlies.