-
Notifications
You must be signed in to change notification settings - Fork 3.1k
[CI] Add bazel TPU presubmit testing #29660
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
base: main
Are you sure you want to change the base?
Conversation
9b4f36d
to
40eccc2
Compare
59d6f13
to
63729cb
Compare
63729cb
to
899aabd
Compare
halt-dispatch-input: ${{ inputs.halt-for-connection }} | ||
- name: Install nightly libtpu | ||
run: | | ||
$JAXCI_PYTHON -m uv pip install --pre libtpu -f https://storage.googleapis.com/jax-releases/libtpu_releases.html |
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.
I think it doesn't affect hermetic Python, e.g. in this case bazel cache will have the LibTPU version defined in the lock file.
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.
IIRC i was getting no libtpu found without it, I'll double check in the morning. I am unsure though if we would want a locked libtpu when it comes to presubmit. Much like XLA this might need the latest nightly or have drift in results. (Both nightly and continuous reference the libtpu nightly)
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.
I'm testing if downloading the LibTPU wheel into the dist
folder (the same one where we have Jax wheels) works.
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.
I haven't installed any wheels at all, the tests passed - https://github.com/jax-ml/jax/actions/runs/16511467263/job/46694360132?pr=30492
Add a new TPU bazel presubmit using 3.13 free threading to increase coverage of our TPU presubmits. This will be introduced as non blocking and be made blocking once reliability is confirmed.
Additionally refactor use of resultstore and rbe cache in jobs that do not use RBE for testing or building. This should allow for use of resultstore and a cache that can allow for testing cache hits.