add nightly builds/tests, remove some cugraph references #61
Conversation
jameslamb
added
improvement
Co-authored-by: Bradley Dice <[email protected]>
| date: ${{ inputs.date }} | ||
| sha: ${{ inputs.sha }} | ||
| script: ci/test_wheel_pylibwholegraph.sh | ||
| matrix_filter: map(select(.ARCH == "amd64")) |
There was a problem hiding this comment.
Just checking - do we know why all these packages skip ARM tests? Should we comment about that?
There was a problem hiding this comment.
I have no idea.
@BradReesWork added that for wholegraph over a year ago: rapidsai/wholegraph#24
@tingyu66 added that constraint for cugraph-dgl and cugraph-pyg about 9 months ago: rapidsai/cugraph#4133
Maybe they or @alexbarghi-nv could share why it's like that.
There was a problem hiding this comment.
There are no packages available on arm64 for pytorch that support CUDA.
There was a problem hiding this comment.
I tried re-enabling those tests and they all failed again because it pulled the pytorch CPU package.
There was a problem hiding this comment.
As far as I know, most ARM64 users of pytorch are going through DLFW or some other means.
There was a problem hiding this comment.
alright thank you. I do see aarch64 CUDA builds of pytorch 2.4.1 (the latest version) at https://anaconda.org/conda-forge/pytorch/files, so maybe in the future there's an opportunity to turn these tests back on.
I'll leave it as-is for the purpose of this PR yet, to get more CI up and running so we can move development over here from the cugraph repo.
|
/merge |
All of this repo's `conda-python-tests` jobs have conditions in them like "skip on ARM": https://github.com/rapidsai/cugraph-gnn/blob/2dd300122dfd6fdea70c9d20c276a3c5946b7613/ci/test_python.sh#L100 https://github.com/rapidsai/cugraph-gnn/blob/2dd300122dfd6fdea70c9d20c276a3c5946b7613/ci/test_python.sh#L141 https://github.com/rapidsai/cugraph-gnn/blob/2dd300122dfd6fdea70c9d20c276a3c5946b7613/ci/test_python.sh#L183 As a result, right now the arm64 `conda-python-tests` jobs are just wasting CI resources... they're spending ~40+~ 5-10 minutes occupying a GPU runner just to download some datasets and then exit ([example build link](https://github.com/rapidsai/cugraph-gnn/actions/runs/11858773988/job/33056063652?pr=69)). This proposes never even starting those jobs, to make CI here less expensive. ## Notes for Reviewers ### But why are we skipping arm at all? Lack of pytorch packages. See #61 (comment) Authors: - James Lamb (https://github.com/jameslamb) Approvers: - Alex Barghi (https://github.com/alexbarghi-nv) - Jake Awe (https://github.com/AyodeAwe) URL: #70
Continues the work of setting up CI here (breaking more changes off of #53).
Proposes adding nightly builds / tests, but not publishing to the nightly PyPI index /
rapidsai-nightlyconda channel. Those publishing jobs should be added here only oncecugraphfolks are ready to fully move development of these libraries to this repo. ref: #58 (comment)This also removes some minor lingering cugraph references that look like they don't belong in this repo.