Who contributed to Google on May 13, 2026?

4 developers shipped this update, including copybara-service[bot], TrevorBergeron, chalmerlowe, and ohmayr.

What were the notable Google updates?

[Mosaic GPU] Implement `CollapseShape` expression to propagate SMEM transforms through `memref.collapse_shape`, [Pallas/Mosaic GPU] Implement lowering for `memref.collapse_shape` and enable `test_wgmma_reshaped_rhs_unit_tiling` under warpgroup semantics, and [pallas-mpmd] Remove `mpmd_meshes` from the lowering context in favor of meshes on avals.

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 REWIRES GPU MEMORY TRANSFORMS; GENAI SDK SHIPS SPEECH TRANSLATION

By RepoJournal · Filed 06:03 UTC on May 13, 2026 · About Google

4 people shipped this

copybara-service[bot] @copybara-service[bot] 4 cited

TrevorBergeron @TrevorBergeron 2 cited

chalmerlowe @chalmerlowe 1 cited

ohmayr @ohmayr 1 cited

JAX's Mosaic GPU stack landed three major commits reshaping how shared memory transforms propagate through tensor operations, while the GenAI SDK reaches v2.2.0 with live speech translation and Agent Platform MCP support.

The JAX team shipped a trio of infrastructure changes that unblock tensor reshaping on GPUs. First, they implemented `CollapseShape` expression handling [1] to propagate SMEM transforms through `memref.collapse_shape`, followed by the full lowering for the operation [2]. These commits fix the test suite for warpgroup semantics—meaning JAX can now efficiently reshape tensors on H100/H200 GPUs without redundant memory copies. In parallel, the pallas-mpmd desk removed `mpmd_meshes` from the lowering context [3], simplifying the abstraction layer for multi-GPU distributed code. The GenAI SDK crossed into v2.2.0 [4] with two feature wins: speech-to-speech translation now works in Gemini Live [5], and Agent Platform MCP support landed with new session and auth helpers [6]. On the cloud-python front, the BigFrames team refactored gbq execution as a semi-executor [7] and introduced `GoogleSqlScalarOp` [8] to improve scalar operation handling, while pandas-gbq got Ruff formatting and mypy type checking [9]. The dependency resolution fix [10] addresses a critical issue in the monorepo—core packages were being fetched remotely instead of from local source, breaking prerelease testing.

Action items

→ Verify JAX GPU tensor reshape operations in your H100 pipelines—these commits enable more efficient memory layouts google/jax [plan]
→ Update python-genai to v2.2.0 if you use Gemini Live speech features googleapis/python-genai [plan]
→ Rebuild nox test sessions in google-cloud-python forks to pull local dependency fixes googleapis/google-cloud-python [immediate]

References

[1] [Mosaic GPU] Implement `CollapseShape` expression to propagate SMEM transforms through `memref.collapse_shape`. ↗ google/jax
[2] [Pallas/Mosaic GPU] Implement lowering for `memref.collapse_shape` and enable `test_wgmma_reshaped_rhs_unit_tiling` under warpgroup semantics. ↗ google/jax
[3] [pallas-mpmd] Remove `mpmd_meshes` from the lowering context in favor of meshes on avals. google/jax
[4] v2.2.0 ↗ googleapis/python-genai
[5] feat: support speech to speech translation in Gemini Live ↗ googleapis/python-genai
[6] chore: Add session and auth helpers for Agent Platform MCP support ↗ googleapis/python-genai
[7] refactor: Unify gbq execution as semi-executor ↗ googleapis/google-cloud-python
[8] refactor(bigframes): Introduce GoogleSqlScalarOp ↗ googleapis/google-cloud-python
[9] feat(pandas-gbq): add Ruff format session and implement mypy ↗ googleapis/google-cloud-python
[10] fix(templates): resolve core dependencies locally and batch pip installs ↗ googleapis/google-cloud-python

Quick answers

What shipped in Google on May 13, 2026?: JAX's Mosaic GPU stack landed three major commits reshaping how shared memory transforms propagate through tensor operations, while the GenAI SDK reaches v2.2.0 with live speech translation and Agent Platform MCP support. In total, 40 commits, 34 pull requests, and 1 releases landed.
Who contributed to Google on May 13, 2026?: 4 developers shipped this update, including copybara-service[bot], TrevorBergeron, chalmerlowe, and ohmayr.
What were the notable Google updates?: [Mosaic GPU] Implement `CollapseShape` expression to propagate SMEM transforms through `memref.collapse_shape`, [Pallas/Mosaic GPU] Implement lowering for `memref.collapse_shape` and enable `test_wgmma_reshaped_rhs_unit_tiling` under warpgroup semantics, and [pallas-mpmd] Remove `mpmd_meshes` from the lowering context in favor of meshes on avals.

JAX FIXES REMAT3 CONSTANT HANDLING CRASH, GOOGLE CLOUD PYTHON CENTRALIZES REST TRANSCODING

Matthew Johnson patched a critical remat3 regression where closed-over constants were silently dropped, causing immediate crashes in production rematerialization pipelines [ref:1].

ai 50 shipped 1-min read

Google 1 day ago

JAX TIGHTENS XLA OPTIMIZATION CONTROLS, GOOGLE CLOUD PYTHON SETS UP FEATURE GATES

JAX makes XLA optimization levels human-readable while Google Cloud Python client libraries prepare a unified feature gating system for observability.

ai 54 shipped 1-min read

Google 2 days ago

GENAI AUDIO TRANSCRIPTION SHIPS, PROTO-PLUS FIXES RACE CONDITIONS

Google's AI SDK gains native audio processing while the core Python cloud libraries ship critical stability fixes across the stack.

ai 60 shipped 1-min read

Google 3 days ago

JAX CUTS ASYNC COLLECTIVE PRIMITIVES, GENAI SDK ADDS LIVE API VOCABULARY

JAX is simplifying its async collective operations while the Python GenAI SDK ships custom vocabulary support for the Live API, marking a shift toward leaner abstractions and richer real-time capabilities.

ai 100 shipped 1-min read

Elsewhere on the wire

AI Agents about 9 hours ago

CLAUDE OPUS 5 LANDS ACROSS THE STACK

The newest Anthropic model is now live in langchain, Cline, and llama-index, with native support for extended reasoning and 1M context windows.

ai-agents 28 shipped 1-min read

Local LLMs about 9 hours ago

OLLAMA LANDS LAGUNA SUPPORT AND CRUSHES MEMORY LEAKS WHILE SGLANG HITS V0.5.16 WITH CONFIDENCE-DRIVEN SPECULATIVE DECODING

Ollama shipped three critical performance and reliability fixes for Metal residency and concurrent access patterns, while SGL-Lang released 0.5.16 with a new speculative algorithm hitting 383.7 tok/s on DeepSeek-V4.