Skip to content
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

CI: test against nightly JAX #1173

Merged
merged 3 commits into from
Jan 15, 2025

Conversation

jakevdp
Copy link
Contributor

@jakevdp jakevdp commented Jan 13, 2025

This will catch potential incompatibilities introduced in JAX before they make it into a release.

Also update parts of the workflow definition to reflect current best practices.

@jakevdp
Copy link
Contributor Author

jakevdp commented Jan 13, 2025

Test failures point to code added earlier today in #1169

@jakevdp
Copy link
Contributor Author

jakevdp commented Jan 13, 2025

Looks like #1172 should fix the failures; I'll rebase on that once it's merged.

@rdyro rdyro self-requested a review January 14, 2025 19:04
@jakevdp
Copy link
Contributor Author

jakevdp commented Jan 14, 2025

Chatting with @rdyro it seems the test failure is probably related to JAX enabling partitionable threefry by default (jax-ml/jax#25363). This change is discussed at jax-ml/jax#18480: This is a new RNG scheme that produces different draws from the same distribution; that change was enough to just exceed the test tolerance with the new random draws.

The best fix is probably to bump the test tolerance.

@jakevdp jakevdp force-pushed the ci-nightly-jax branch 3 times, most recently from d02396b to dfcfc98 Compare January 15, 2025 01:39
@copybara-service copybara-service bot merged commit 98c73c5 into google-deepmind:main Jan 15, 2025
12 checks passed
@jakevdp jakevdp deleted the ci-nightly-jax branch January 15, 2025 20:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants