### Bug Fixes and Other Changes

*   Add TensorFlow Quantizer to TensorFlow pip package.

*   `tf.sparse.segment_sum` `tf.sparse.segment_mean` `tf.sparse.segment_sqrt_n` `SparseSegmentSum/Mean/SqrtN[WithNumSegments]`

    *   Added `sparse_gradient` option (default=false) that makes the gradient of these functions/ops sparse (`IndexedSlices`) instead of dense (`Tensor`), using new `SparseSegmentSum/Mean/SqrtNGradV2` ops.

- cloud.google.com/go/datastore: v1.0.0
- cloud.google.com/go/pubsub: v1.0.1
- cloud.google.com/go/storage: v1.0.0
- dmitri.shuralyov.com/gpu/mtl: 666a987
- github.com/cespare/xxhash/v2: [v2.1.1](https://github.com/cespare/xxhash/v2/tree/v2.1.1)
- github.com/checkpoint-restore/go-criu/v4: [v4.0.2](https://github.com/checkpoint-restore/go-criu/v4/tree/v4.0.2)

pkg syscall (freebsd-arm64), type Stat_t struct, Rdev uint64
pkg syscall (freebsd-arm64), type Stat_t struct, Size int64
pkg syscall (freebsd-arm64), type Stat_t struct, Spare [10]uint64
pkg syscall (freebsd-arm64), type Stat_t struct, Uid uint32
pkg syscall (freebsd-arm64), type Statfs_t struct
pkg syscall (freebsd-arm64), type Statfs_t struct, Asyncreads uint64

pkg syscall (freebsd-riscv64), type Stat_t struct, Rdev uint64 #53466
pkg syscall (freebsd-riscv64), type Stat_t struct, Size int64 #53466
pkg syscall (freebsd-riscv64), type Stat_t struct, Spare [10]uint64 #53466
pkg syscall (freebsd-riscv64), type Stat_t struct, Uid uint32 #53466
pkg syscall (freebsd-riscv64), type Statfs_t struct #53466
pkg syscall (freebsd-riscv64), type Statfs_t struct, Asyncreads uint64 #53466

        * [Better with `Annotated`](https://fastapi.tiangolo.com/tutorial/path-params-numeric-validations/#better-with-annotated)
    * [Dependencies - First Steps - Share `Annotated` dependencies](https://fastapi.tiangolo.com/tutorial/dependencies/#share-annotated-dependencies)

width w instead of // radix 16. This is like a binary representation (one digit // for each binary place) but we allow the digits to grow in // magnitude up to 2^{w-1} so that the nonzero digits are as // sparse as possible. Intuitively, this "condenses" the // "mass" of the scalar onto sparse coefficients (meaning // fewer additions). basepointNafTable := basepointNafTable() var aTable nafLookupTable5 aTable.FromP3(A) // Because the basepoint is fixed, we can use a wider NAF // corresponding to...

width w instead of // radix 16. This is like a binary representation (one digit // for each binary place) but we allow the digits to grow in // magnitude up to 2^{w-1} so that the nonzero digits are as // sparse as possible. Intuitively, this "condenses" the // "mass" of the scalar onto sparse coefficients (meaning // fewer additions). basepointNafTable := basepointNafTable() var aTable nafLookupTable5 aTable.FromP3(A) // Because the basepoint is fixed, we can use a wider NAF // corresponding to...