-
Notifications
You must be signed in to change notification settings - Fork 165
CI: Add Windows GPU runner for tests #444
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
Conversation
/ok to test |
This comment has been minimized.
This comment has been minimized.
/ok to test |
@cryos @rwgk do you guys have clever ways about refactoring a workflow file? On the Windows runner, we cannot launch any container so this part must be removed: cuda-python/.github/workflows/test-wheel.yml Lines 32 to 38 in 7387715
but otherwise the remaining logic should be the same and preserved... Are there smarter ways than duplicating the file for Windows? |
I feel like there should be, searching turned up the same disappointing you cannot make |
/ok to test |
1 similar comment
/ok to test |
/ok to test |
(spoke to George, the underlying image is changed now, let me restart the tests) |
/ok to test |
/ok to test |
Looks like we hit https://github.com/orgs/community/discussions/138813... which is weird... I haven't seen this error in the previous project... |
Unfortunately there is no way to route this issue to the right GitHub support person. Closing this for now. |
/ok to test |
/ok to test |
/ok to test |
/ok to test |
/ok to test |
/ok to test |
This is ready for review! The PR description is updated. |
The doc build failure
is due to Sphinx 8.2.0, let's fix it in a separate PR: |
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.
Very cool, looks good to me.
|
Part of #277.
Based on offline discussion with our admin @aflat and @aterrel @kkraus14, as a stop-gap solution until we get our own runners we are setting up a GH-hosted runners to cover Windows. Because the rate is high, we pick only 1 Python version x 2 latest CUDA major versions (11.8 & 12.8) x 2 different ways of getting CUDA to keep the cost within $100 a month.
In terms of the implementation, the new workflow is largely the same as its Linux counterpart, except for the following changes since the VM image of choice (Windows 11 23H2) has nothing pre-installed:
fetch_ctk
action cannot be used; instead, we switch to useJimver/cuda-toolkit
which has been approved by the admins