Who contributed to Google on August 4, 2026?

8 developers shipped this update, including copybara-service, Mark Sandler, Annie Luc, Mark Daoust, Sakthivel Subramanian, Chalmer Lowe, Heba Alazzeh, and release-please[bot].

What were the notable Google updates?

[Mosaic GPU] Add support for transposed 8-bit stores (using ld/stmatrix), [Mosaic GPU] Support print_layout for SMEM refs, and Exclude drop_fields fields when auto-populating data/meta fields.

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 HARDENS GPU KERNELS WHILE CLOUD PYTHON PATCHES SUPPLY CHAIN RISK

By RepoJournal · Filed 06:59 UTC on August 4, 2026 · About Google

8 people shipped this

copybara-service @copybara-service 4 cited

Mark Sandler @marksandler2 1 cited

Annie Luc @Annhiluc 1 cited

Sakthivel Subramanian @sakthivelmanii 1 cited

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

Mark Daoust @MarkDaoust

Chalmer Lowe @chalmerlowe

Heba Alazzeh @hebaalazzeh

JAX shipped critical Mosaic GPU optimizations that squeeze performance out of transposed memory transfers, while the cloud SDK locked down six open security vulnerabilities in sqlalchemy-spanner.

JAX's Mosaic GPU layer landed transposed 8-bit store support [1], a non-trivial integration that bundles 8x16 transfers into 16x16 operations to work within half-warp constraints. The same subsystem gained print_layout support for shared memory references [2], adding observability into tiling and swizzle patterns. On the dataclass front, JAX relaxed register_dataclass constraints [3] to exclude drop_fields from auto-population, eliminating brittle inheritance boilerplate that plagued users. Pallas saw two critical fixes: the memory space logic for mpmd_map now mirrors default_mesh_discharge_rule [4], preparing the eventual removal of core_map, and copy_gmem_to_smem no longer mandates barriers on pre-Hopper GPUs [5], exposing functionality already available in Mosaic GPU. Over in google-cloud-python, sqlalchemy-spanner pulled in pyasn1 0.6.4, urllib3 2.7.0, and idna 3.18 [7] to resolve six open Dependabot security alerts, while bigframes shipped v2.47.0 [8] with new AI embedding and similarity functions for BigQuery accessors. The python-genai SDK saw routine maintenance including deprecated field cleanup [6].

Action items

→ If using sqlalchemy-spanner, upgrade to pull in the patched transitive dependencies immediately googleapis/google-cloud-python [immediate]
→ Review your Pallas multi-GPU kernels for unnecessary barriers on pre-Hopper hardware and remove them google/jax [plan]
→ Update bigframes to 2.47.0 if you need AI embedding support in BigQuery googleapis/google-cloud-python [plan]

References

[1] [Mosaic GPU] Add support for transposed 8-bit stores (using ld/stmatrix) ↗ google/jax
[2] [Mosaic GPU] Support print_layout for SMEM refs. ↗ google/jax
[3] Exclude drop_fields fields when auto-populating data/meta fields. google/jax
[4] [pallas] Mirrored the `default_mesh_discharge_rule` memory space logic for `mpmd_map` ↗ google/jax
[5] [pallas:mgpu] `copy_gmem_to_smem` no longer requires a barrier on pre-Hopper GPUs ↗ google/jax
[6] chore: Cleanup deprecated field response_modalities googleapis/python-genai
[7] chore(sqlalchemy-spanner): update vulnerable transitive dependencies in sqlalchemy-spanner (#17971) googleapis/google-cloud-python
[8] bigframes: v2.47.0 ↗ googleapis/google-cloud-python

Quick answers

What shipped in Google on August 4, 2026?: JAX shipped critical Mosaic GPU optimizations that squeeze performance out of transposed memory transfers, while the cloud SDK locked down six open security vulnerabilities in sqlalchemy-spanner. In total, 52 commits, 47 pull requests, and 1 releases landed.
Who contributed to Google on August 4, 2026?: 8 developers shipped this update, including copybara-service, Mark Sandler, Annie Luc, Mark Daoust, Sakthivel Subramanian, Chalmer Lowe, Heba Alazzeh, and release-please[bot].
What were the notable Google updates?: [Mosaic GPU] Add support for transposed 8-bit stores (using ld/stmatrix), [Mosaic GPU] Support print_layout for SMEM refs, and Exclude drop_fields fields when auto-populating data/meta fields.

JAX PINS PYTHON 3.14T, GENAI GAINS ROBOTICS MODEL AND ERROR HANDLING

JAX disabled profiler tests on free-threaded Python to unblock CI, while python-genai 2.17.0 ships the Gemini Robotics ER 2 Preview and fixes tool call limits.

ai 90 shipped

Google Aug 5

JAX ADDS NUMPY 2.6 TOP_K, INTEL GPU SUPPORT LANDS IN CI

JAX shipped native top_k alignment with NumPy 2.6.0 while simultaneously landing full OneAPI CI infrastructure for Intel GPU testing across the entire stack.

ai 59 shipped

Google Aug 3

JAX SYNCS XLA DEPENDENCY TO LATEST REVISION

JAX pinned its XLA dependency to a fresh upstream revision overnight, pulling in compiler optimizations and fixes across the ML stack.

ai 1 shipped

Google Aug 2

JAX AVERTS BREAKING CHANGE IN CUSTOM_GRADIENT API

JAX pulled back from making the `fun` parameter positional-only in custom_gradient, preserving backward compatibility just before release.

ai 3 shipped

Elsewhere on the wire

AI Agents Aug 6

LANGCHAIN HARDENS ANTHROPIC TOOL CALLING AND GIT CREDENTIALS AS OPENHANDS SHIPS 1.10.0

LangChain fixed a critical credential leak in evaluation runs while Anthropic tool schemas get proper fallback handling, as three major agent frameworks push releases across the ecosystem.

ai-agents 54 shipped

Local LLMs Aug 6

OLLAMA FIXES KV CACHE CORRUPTION, LLAMA.CPP HARDENS FILE SEARCH, VLLM KILLS MOE LEGACY CODE

Ollama patched a critical cache corruption bug that corrupts lazy snapshots during multi-request prefills, while llama.cpp hardened Windows junction traversal and vLLM removed 14 hardcoded MoE shards across quantization methods.