vLLM
Tests for vLLM's reasoning-content mirror, merged; found while serving GLM-5.3 on one GPU.
Open source
Merged upstream
Most of the work is in Keras and KerasHub; the rest is fixes across the Python and ML tooling people use every day.
Keras and KerasHub
The larger pieces, each with its own page.
- Keras on PyTorch: compiled generation 11.6x fasterJul 2026 – Oct 2026
Traced why Keras on PyTorch ran about 9x slower than plain PyTorch, and merged the fixes upstream, one root cause each. · Google · Merged
- One call from a Keras model to a phone: LiteRT exportDec 2025 – Jul 2026
Built the LiteRT export path in Keras and KerasHub: a trained model to a phone-ready file in one call, on TensorFlow and PyTorch. · Google · Merged
- Llama 3.1 in KerasHub, and Gemma 3 fixesApr 2025 – Jul 2026
Implemented Llama 3.1 in KerasHub and fixed Gemma 3 bugs, including a preprocessor crash across Gemma 3, 3n and 4. · Google · Merged
- Keras correctness fixesJun 2025 – Oct 2026
Fixed TPU attention, N-D attention inputs, nested weight saving, masks, call-context leaks and ONNX export of models over 1.5 GB. · Google · Merged
- Making KerasHub models exportable
Fixed the layers that kept KerasHub models such as ESM, PARSeq and Flux from exporting to LiteRT. · Google · Merged
Elsewhere
Python tooling fixes
Merged fixes in pip, pydantic, pyright, typeshed, poetry, trio and pytest plugins.
Built and proposed, not merged
Larger designs that went upstream as proposals and were not merged.
- Model-parallel layout maps for KerasHub LLMsJul 2026
Designed model-parallel layout maps for KerasHub LLMs, from Gemma and Qwen to Llama and Mixtral; proposed upstream, not merged. · Google · Proposed, not merged
- LiteRT-LM export for KerasHub LLMs
Built one call from a KerasHub LLM to a LiteRT-LM bundle with tokenizer and prefill and decode signatures; proposed upstream, not merged. · Google · Proposed, not merged