Who contributed to Google on August 13, 2026?

7 developers shipped this update, including hawkinsp, copybara-service, Google Team Member, Annie Luc, release-please[bot], Daniel Sanche, and ohmayr.

What were the notable Google updates?

Switch to bzlmod by default, Implement AOT compilation for autotuned JAX-Triton kernels, and Regenerate requirements lock files.

Google

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

Pick a date

Topics: AI / ML Python Full archive →

The Wire · Showcase

JAX SWITCHES TO BZLMOD, SHIPS AOT COMPILATION FOR TRITON KERNELS

By RepoJournal · Filed 06:59 UTC on August 13, 2026 · About Google

7 people shipped this

copybara-service @copybara-service 4 cited

hawkinsp @hawkinsp 2 cited

ohmayr @ohmayr 2 cited

Google Team Member @google-genai-bot 1 cited

Annie Luc @Annhiluc 1 cited

release-please[bot] @release-please[bot] 1 cited

Daniel Sanche @daniel-sanche 1 cited

JAX's build system migration to Bzlmod is live, paired with major performance improvements for autotuned GPU kernels that compile at instantiation rather than runtime.

The JAX team switched to Bzlmod by default [1], a significant infrastructure move that restructures how pip dependencies are exposed in the build system. Pip packages now route through `@pypi//<package>` rather than standalone repositories, requiring updates to lockfiles and dependency declarations. Simultaneously, JAX shipped AOT compilation for autotuned Triton kernels [2], moving compilation of candidate configurations from JIT time into the instantiate phase. This eliminates serialization overhead since instantiate and execute now happen in the same process. Supporting infrastructure landed across three commits: Triton now handles host-side TMA descriptors [4], the pallas multi-GPU layer auto-infers `PROMISE_IN_BOUNDS` in pipeline emission [5], and requirements lock files were regenerated [3] to accommodate the Bzlmod transition. On the GenAI front, the Python SDK cleaned deprecated Turn types [7], dropped unused auth imports [6], and fixed automatic function calling bugs in streaming contexts [8]. BigFrames hit v2.48.0 with expanded string operation support in the transpiler [9], while BigTable standardized client-side metrics across implementations [10]. The google-cloud-python repo saw two reverts [11] [12] of generator and librarian changes, suggesting recent commits needed correction.

Action items

→ Review JAX Bzlmod migration impact on your build pipeline if using JAX internally google/jax [plan]
→ Test JAX AOT compilation changes if using Triton kernels with autotuning google/jax [monitor]
→ Update to BigFrames v2.48.0 if you depend on string transpilation googleapis/google-cloud-python [plan]
→ Verify BigTable client-side metrics are aligned with your monitoring expectations googleapis/google-cloud-python [monitor]

References

[1] Switch to bzlmod by default. ↗ google/jax
[2] Implement AOT compilation for autotuned JAX-Triton kernels. ↗ google/jax
[3] Regenerate requirements lock files. ↗ google/jax
[4] [triton] Added support for host-side TMA descriptors ↗ google/jax
[5] [pallas:mgpu] Automatically infer `PROMISE_IN_BOUNDS` in `emit_pipeline` ↗ google/jax
[6] chore: drop unused auth imports from the MCP utils module googleapis/python-genai
[7] chore: remove deprecated Turn types googleapis/python-genai
[8] fix: improve AFC(automatic function calling) in chat including fixing bugs in AFC for generate_content_stream. also log warnings in generate_content, generate_content_stream and their async variants t ↗ googleapis/python-genai
[9] bigframes: v2.48.0 ↗ googleapis/google-cloud-python
[10] fix(bigtable): standardize client side metrics (#17899) googleapis/google-cloud-python
[11] Revert "fix(generator): use flat_ref_types in test templates and delete remove-unused-imports" (#18088) googleapis/google-cloud-python
[12] Revert "chore(librarian): remove post processing script for ces" (#18094) googleapis/google-cloud-python

Quick answers

What shipped in Google on August 13, 2026?: JAX's build system migration to Bzlmod is live, paired with major performance improvements for autotuned GPU kernels that compile at instantiation rather than runtime. In total, 49 commits, 44 pull requests, and 2 releases landed.
Who contributed to Google on August 13, 2026?: 7 developers shipped this update, including hawkinsp, copybara-service, Google Team Member, Annie Luc, release-please[bot], Daniel Sanche, and ohmayr.
What were the notable Google updates?: Switch to bzlmod by default, Implement AOT compilation for autotuned JAX-Triton kernels, and Regenerate requirements lock files.

JAX LOCKS IN PYTHON 3.15 SUPPORT AS GENAI SDK SHIPS VIDEO ENHANCEMENTS

JAX is ready for Python 3.15 and its free-threaded runtime, while Google's Genai SDK adds video resolution controls and BigQuery gets Apache Arrow query results.

ai 95 shipped

Google Aug 12

JAX DROPS PYTHON 3.13T SUPPORT, CLEANS UP BUILD PIPELINE

JAX consolidated its dependency lock files after dropping Python 3.13 free-threading support, signaling a shift in the framework's build strategy as it pushes through infrastructure modernization.

ai 91 shipped

Google Aug 11

JAX WIRES INTEL GPU SUPPORT, FIXES MOSAIC ASSEMBLY BUGS

JAX landed native random number generation on Intel GPUs while fixing critical debugging directives in Mosaic GPU inline assembly, signaling accelerated support across vendor platforms.

ai 74 shipped

Google Aug 10

JAX REVERTS BREAKING CHANGE, SYNCS XLA UPSTREAM

A changelist that hit production got rolled back twice in 12 hours, signaling either a critical regression or a coordination miss that JAX caught before it spread.

ai 3 shipped

Elsewhere on the wire

AI Agents Aug 14

LANGCHAIN SHIPS OPENAI 3.0 SUPPORT, CLINE PATCHES SANDBOX LEAKS

LangChain's openai integration now supports the OpenAI Python SDK 3.0 while maintaining backward compatibility, and Cline fixes unbounded subprocess growth in long-running deployments.

ai-agents 37 shipped

Local LLMs Aug 14

OLLAMA LAUNCHES DEEPSEEK AND MUSE INTEGRATIONS WHILE VLLM SHIPS HARDWARE-AGNOSTIC MODELING

Ollama just opened its launch ecosystem to DeepSeek Harness and Meta's Muse Code, while vLLM pivots to HuggingFace transformers as a unified backend for tail models.

+12

llms 261 shipped

@CachyOS Aug 14

CACHY AUR KEEPS PACE: FOUR MAJOR PACKAGE BUMPS OVERNIGHT

Claude, ChatGPT, and Zen Browser all shipped point releases into the AUR-derived repos while the core PKGBUILDS pipeline handled Go, Python, and OBS Studio rebuilds.