Who contributed to Google on May 15, 2026?

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

What were the notable Google updates?

Remove constructor arguments in JaxprEqnContext and always read the value of context by default, [internal] move random-related sources into a new `jax._src.random` submodule, and Add compiler_options to compute_on and add tests for sparsecore offload of collectives.

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 CLEANS HOUSE WITH CONSTRUCTOR SIMPLIFICATION AND RANDOM MODULE REORGANIZATION

By RepoJournal · Filed 06:03 UTC on May 15, 2026 · About Google

4 people shipped this

copybara-service[bot] @copybara-service[bot] 6 cited

shuoweil @shuoweil 1 cited

sycai @sycai 1 cited

TrevorBergeron @TrevorBergeron 1 cited

JAX stripped JaxprEqnContext down to its essentials, removing constructor arguments in favor of context-by-default reading, while consolidating all PRNG logic into a unified namespace to handle the growing codebase.

The JaxprEqnContext refactor [1] cuts surface area and reduces initialization complexity, a pattern that pairs cleanly with JAX's broader internal reorganization. The team moved random-related sources into jax._src.random [2], keeping the public API unchanged while grouping all PRNG code in one place as the feature set grows. On the compilation front, compute_on gained compiler_options support and now tests sparsecore offload of collectives [3], with legacy compute_on tests replaced by compute_on2 [4] for cleaner test coverage. Internal cleanup continued across lax with trace_to_jaxpr_dynamic replaced by trace_to_jaxpr [5], simplifying the callable interface without touching user-visible behavior. Meanwhile, the Google Cloud Python ecosystem made a hard choice: dropping Python 3.9 and earlier across generated packages [8], resolving dependency conflicts and removing EOL version baggage. BigQuery DataFrames completed the deprecation of experimental Multimodal Blob APIs [9] while BigFrames added output_mode support for ai.classify [10], and executor internals went async for better concurrency [11]. The python-genai SDK shipped refreshed docs for 2.2 [6] and added output shortcut properties to interaction objects [7], letting developers skip the steps timeline dance.

Action items

→ Review JAX internal imports if using trace_to_jaxpr_dynamic directly google/jax [plan]
→ Plan Python 3.9 EOL migration for google-cloud-python dependencies googleapis/google-cloud-python [plan]
→ Update Multimodal Blob code to use _from_glob_path before next release googleapis/google-cloud-python [plan]
→ Test python-genai 2.2 interaction output shortcuts in dev environment googleapis/python-genai [monitor]

References

[1] Remove constructor arguments in JaxprEqnContext and always read the value of context by default. google/jax
[2] [internal] move random-related sources into a new `jax._src.random` submodule. ↗ google/jax
[3] Add compiler_options to compute_on and add tests for sparsecore offload of collectives. ↗ google/jax
[4] Replace compute_on tests with compute_on2 ↗ google/jax
[5] jax/lax: replace trace_to_jaxpr_dynamic with trace_to_jaxpr ↗ google/jax
[6] docs: refresh generated docs for 2.2 ↗ googleapis/python-genai
[7] feat: interaction.{output_text,output_image,output_audio,output_video} ↗ googleapis/python-genai
[8] feat: regenerate google-cloud-a packages (#17089) googleapis/google-cloud-python
[9] feat!: complete deprecation and cleanup of multimodal blob APIs ↗ googleapis/google-cloud-python
[10] feat(bigframes): support output_mode for ai.classify ↗ googleapis/google-cloud-python
[11] refactor(bigframes): make executor internals async ↗ googleapis/google-cloud-python

Quick answers

What shipped in Google on May 15, 2026?: JAX stripped JaxprEqnContext down to its essentials, removing constructor arguments in favor of context-by-default reading, while consolidating all PRNG logic into a unified namespace to handle the growing codebase. In total, 44 commits and 42 pull requests landed.
Who contributed to Google on May 15, 2026?: 4 developers shipped this update, including copybara-service[bot], shuoweil, sycai, and TrevorBergeron.
What were the notable Google updates?: Remove constructor arguments in JaxprEqnContext and always read the value of context by default, [internal] move random-related sources into a new `jax._src.random` submodule, and Add compiler_options to compute_on and add tests for sparsecore offload of collectives.

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.