Conversation
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
5d946e4 to
8682c2e
Compare
meta-cla
bot
added
the
CLA Signed
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
8682c2e to
0c9c085
Compare
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
0c9c085 to
d7cf4f9
Compare
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
d7cf4f9 to
ed5d5f1
Compare
oulgen
left a comment
oulgen
left a comment
There was a problem hiding this comment.
I meant test ci not lint ci
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
ed5d5f1 to
645e8b3
Compare
aditvenk
changed the title
Add a lint-macos job that runs pre-commit on the macos-m2-26 self-hosted runner. Installs PyTorch from the CPU/MPS nightly index (no CUDA on macOS) and uses [dev,pallas-cpu] extras (no cute-cu12). stack-info: PR: #1859, branch: aditvenk/stack/23
aditvenk
force-pushed
the
aditvenk/stack/23
branch
from
645e8b3 to
f83bd0f
Compare
Linting on macOS can have failures not seen on Linux due to non-availability of certain packages like Triton/Cutlass. Running a lint job on mac is very quick (< 20s), so I'd suggest to keep it. I have modified the approach to not introduce a new yml, and instead modify lint.yml. The change to matrix.json & test.yml is in the future PR in the stack: #1862 |
oulgen
left a comment
oulgen
left a comment
There was a problem hiding this comment.
I dont think we need lint for macos. It is a subset of the full lint
If we were to add a dependency on some other package that does not exist on Mac, then we would need to update the ignore modules list. Having it in CI will catch such issues. But I can close this Pr if we are okay if finding such issues lazily |
|
Closing. We can add lint job for mac in the future if needed. Test CI tracked by #1862 |
| fi | ||
|
|
||
| lint-macos: | ||
| runs-on: macos-m2-26 |
There was a problem hiding this comment.
If you'll need to run anything, please take a stock macos, or at the very least macos-m2-15, because there are only few 26 out there
There was a problem hiding this comment.
I'd prefer that too, but unfortunately MetalPerformancePrimitives (MPP) which will be the lowering target for matmul, reduction (in future PRs) is available only from macos-26
4 participants
Stacked PRs:
Add macOS lint job to lint.yml
Add a lint-macos job that runs pre-commit on the macos-m2-26
self-hosted runner. Installs PyTorch from the CPU/MPS nightly index
(no CUDA on macOS) and uses [dev,pallas-cpu] extras (no cute-cu12).