round robin reads, users can guarantee that consumers get similar-sized
        examples in the same step.
    *   tf.data service now supports optional compression. Previously data would
        always be compressed, but now you can disable compression by passing
        `compression=None` to `tf.data.experimental.service.distribute(...)`.
    *   `tf.data.Dataset.batch()` now supports `num_parallel_calls` and

        * Run middleware-like code only for a subset of *path operations*.
        * Process a request before passing it to a *path operation function*. E.g. decompressing, deserializing, etc.
        * Processing a response after being generated by *path operation functions* but before returning it. E.g. adding custom headers, logging, adding extra metadata.

privBytes[:32]) hs := sha512.New() hs.Write(priv.seed[:]) h := hs.Sum(make([]byte, 0, sha512Size)) if _, err := priv.s.SetBytesWithClamping(h[:32]); err != nil { panic("ed25519: internal error: setting scalar failed") } // Note that we are not decompressing the public key point here, // because it takes > 20% of the time of a signature generation. // Signing doesn't use it as a point anyway. copy(priv.pub[:], privBytes[32:]) copy(priv.prefix[:], h[32:]) if err := fipsPCT(priv); err != nil { // This...