* Introducing `tf.types.experimental.AtomicFunction` as the fastest way to perform TF computations in Python.

    * Can be accessed through `inference_fn` property of `ConcreteFunction`s

  - `spec.backend` -> `spec.defaultBackend`
  - `serviceName` -> `service.name`

v1.0.0" "crypto/internal/fips140" "crypto/internal/sysrand" "io" "sync" "sync/atomic" ) // memory is a scratch buffer that is accessed between samples by the entropy // source to expose it to memory access timings. // // We reuse it and share it between Seed calls to avoid the significant (~500µs) // cost of zeroing a new allocation every time. The entropy source accesses it // using atomics (and doesn't care about its contents). // // It should end up in the .noptrbss section, and become backed...