overhead in those modes. // To bring the old performance back, we mark all functions that loop // over Nat words with //go:norace. Because //go:norace does not // propagate across inlining, we must also mark functions that inline // //go:norace functions - specifically, those that inline add, addMulVVW, // assign, cmpGeq, rshift1, and sub. // choice represents a constant-time boolean. The value of choice is always // either 1 or 0. We use an int instead of bool in order to make decisions in // constant...

* ⬆ [pre-commit.ci] pre-commit autoupdate. PR [#13656](https://github.com/fastapi/fastapi/pull/13656) by [@pre-commit-ci[bot]](https://github.com/apps/pre-commit-ci).
* ✅ Use `inline-snapshot` to support different Pydantic versions in the test suite. PR [#12534](https://github.com/fastapi/fastapi/pull/12534) by [@15r10nk](https://github.com/15r10nk).

overhead in those modes. // To bring the old performance back, we mark all functions that loop // over Nat words with //go:norace. Because //go:norace does not // propagate across inlining, we must also mark functions that inline // //go:norace functions - specifically, those that inline add, addMulVVW, // assign, cmpGeq, rshift1, and sub. // choice represents a constant-time boolean. The value of choice is always // either 1 or 0. We use an int instead of bool in order to make decisions in // constant...

*   `saved_model.utils` now support SparseTensors transparently.
*   A more efficient implementation of non-max suppression.
*   Add support for the shrinkage-type L2 to FtrlOptimizer in addition to the
    online L2 it already supports.
*   Fix negative variance in moments calculation.
*   Expand UniqueOp Benchmark Tests to cover more collision cases.
*   Improves stability of GCS filesystem on Mac.