RepoJournal
Google

Google

JAX, the GenAI SDK, and the Cloud libs - Google's open source layer

Keep up with Google in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

One email a day. Unsubscribe in one click.

Pick a date

Topics: AI / ML Python Full archive →

The Wire · Showcase

JAX CLEANS UP CI PIPELINE, GOOGLE CLOUD PYTHON BATTLES TEST FLAKES

By RepoJournal · Filed · About Google

JAX is splitting its monolithic GitHub Actions workflow into four focused pipelines while standardizing job names across the entire test matrix [ref:1], the same week Google Cloud Python squashes a cascade of flaky Firestore tests and pandas 3 incompatibilities.

The JAX repo is doing infrastructure theater that actually matters: renaming job identifiers for clarity, collapsing free-threaded Python from `3.14-nogil` to `3.14t`, and splitting one unwieldy workflow into separate test, lint, security, and docs pipelines [1]. This reduces cognitive load on every PR review and makes CI signals faster to read. On the language side, JAX shipped a significant upgrade to its effects system by replacing de Bruijn indices with explicit variables [2], which matters for anyone building custom gradient rules or composing advanced autodiff patterns. The Mosaic TPU crew deprecated `tpu.rotate()` in favor of `tpu.dynamic_rotate()`, which now handles both static and dynamic rotation in one call [3], and CUDA support now extends to pinned host memory tensors via DLPack, unblocking PyTorch interop workflows [4]. Over in Google Cloud Python, the team is burning down test infrastructure debt: Firestore's flaky suite is fixed by switching to 10 worker threads instead of auto-scaling [9], pandas 3 failures are patched [10], and redundant named database test sweeps are eliminated to cut execution time . The python-genai repo is quietly shipping GAOS SDK library integrations across multiple PRs [5][6][7], and a new workflow to flag stale google-contributor PRs just landed [8].

One email a day. Unsubscribe in one click.

Action items

References

  1. [1] Clean up GitHub Actions job names for presubmits. google/jax
  2. [2] Use vars instead of de bruijn indices in effects ↗ google/jax
  3. [3] [Mosaic TPU] Deprecated `tpu.rotate`. Use `tpu.dynamic_rotate` instead, which supports both static and dynamic rotate. google/jax
  4. [4] [dlpack] [cuda] Support pinned host tensors in from_dlpack ↗ google/jax
  5. [5] feat: introduce gaos sdk libs ↗ googleapis/python-genai
  6. [6] feat: introduce gaos sdk libs ↗ googleapis/python-genai
  7. [7] feat: introduce gaos sdk libs ↗ googleapis/python-genai
  8. [8] chore: Add GH workflow for stale google-contributor PRs googleapis/python-genai
  9. [9] chore(firestore): fix firestore flaky tests ↗ googleapis/google-cloud-python
  10. [10] chore: address pandas 3 failure and remove inherently flaky system test (#17452) googleapis/google-cloud-python

Quick answers

What shipped in Google on June 13, 2026?
JAX is splitting its monolithic GitHub Actions workflow into four focused pipelines while standardizing job names across the entire test matrix [ref:1], the same week Google Cloud Python squashes a cascade of flaky Firestore tests and pandas 3 incompatibilities. In total, 39 commits and 34 pull requests landed.
Who contributed to Google on June 13, 2026?
4 developers shipped this update, including dougalm, gspschmid, copybara-service[bot], and ohmayr.
What were the notable Google updates?
Clean up GitHub Actions job names for presubmits, Use vars instead of de bruijn indices in effects, and [Mosaic TPU] Deprecated `tpu.rotate`. Use `tpu.dynamic_rotate` instead, which supports both static and dynamic rotate.

More from Google

Daily updates, in your inbox

Follow Google

Keep up with Google in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

One email a day. Unsubscribe in one click. Read a past issue →

Elsewhere on the wire

Want every project, not just this one?

We use privacy-friendly analytics (Google Analytics, IP-anonymized) to see which pages help readers. No ads, and we never sell your data. See our Privacy Policy.