ci: pin numpy below 2.5 in the GPU test environment - #4313
Merged
Conversation
The GPU test jobs fail on every PR since the conda solve started picking numpy 2.5.2: numba-cuda 0.30.4 registers an overload for np.row_stack unconditionally for numpy >= 2.0, but numpy 2.5 removed np.row_stack, so importing numba_cuda's numpy support raises AttributeError and the tests-cuda suite fails. Constrain numpy to <2.5 in the micromamba create-args (where the CI environment is solved) and in requirements-test-gpu.txt (for manual pip installs). Drop the pin when a numba-cuda release supports numpy 2.5 (NVIDIA/numba-cuda#907); the candidate fix is NVIDIA/numba-cuda#916, unmerged as of this commit. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ariostas
approved these changes
Aug 21, 2026
ariostas
left a comment
ariostas
left a comment
Member
There was a problem hiding this comment.
Thank you, @TaiSakuma! Yeah, we had to do the same for uproot
|
The documentation preview is ready to be viewed at http://preview.awkward-array.org.s3-website.us-east-1.amazonaws.com/PR4313 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The
Run GPU Testsjobs fail on every PR since about 2026-08-15 (currently #4294 and #4312). The GPU environment is solved fresh from conda-forge with numpy unconstrained, and the solver now picks numpy 2.5.2. numba-cuda 0.30.4 registers an overload fornp.row_stackunconditionally for numpy >= 2.0, but numpy 2.5 removednp.row_stack(deprecated since 2.0), so importingnumba_cuda/numba/cuda/np/npyimpl.pyraisesAttributeError: module 'numpy' has no attribute 'row_stack'and thetests-cudasuite fails.This PR constrains numpy to below 2.5 in two places:
.github/actions/setup-gpu-env/action.yml: added to the micromambacreate-args, so the conda solver picks a consistent environment. This is the change that fixes CI.requirements-test-gpu.txt: the same constraint for manual pip installs, which hit the same failure because the numba-cuda wheel does not cap numpy either. In CI this line is a no-op: the pip step finds the constraint already satisfied by the conda-installed numpy.The pin is temporary. numba-cuda tracks numpy 2.5 support in NVIDIA/numba-cuda#907, and the candidate fix (NVIDIA/numba-cuda#916, mirroring the guard that numba 0.67.0 already ships) is unmerged as of this PR; the latest numba-cuda release, v0.30.4, predates it. Both files carry a TODO comment naming this removal condition.
After this merges, a CI re-run on the affected PRs picks up the fix without a rebase, because the
pull_requestmerge ref includes main.🤖 Generated with Claude Code