September 2025 highlights for google/flax: Focused on performance profiling, API modernization, and reliability improvements. Delivered FLOPs reporting in tabulate, introduced standalone public APIs for iter_modules/iter_children with a deprecation path for legacy Module methods, and strengthened module tree integrity and VJP correctness to prevent double counting and handle shared structures. Also improved tests and typing hygiene and updated documentation to reflect API changes and deprecation strategy. Impact: clearer cost-aware analysis for forward/backward passes, safer API evolution, and higher reliability of VJP/tabulation workflows.

August 2025 highlights for pytorch/pytorch: focused feature delivery with accompanying tests and bindings to improve usability and portability. Key work included two major feature enhancements: (1) Stable tensor API: added is_cpu method with tests and Python bindings; (2) Stable ABI: ported amax operation for torchaudio with single- and vectorized implementations and tests. These changes enhance runtime diagnostics, stabilize interfaces for downstream consumers, and improve cross-repo stability. No major bugs fixed this month; emphasis was on delivering robust features and validating through comprehensive tests.