// the handle loses information, but `SummarizeValue` would be more precise.
  virtual bool PreferCustomSummarizer() const { return false; }

  // Returns a string which summarizes the value of this TensorHandle, for
  // debugging. Does not include a shape or dtype.
  //
  // Included in the default implementation of DebugString.
  virtual absl::Status SummarizeValue(std::string& summary) const;

  ImmediateExecutionTensorHandle* unwrapped_handle =
      tensorflow::unwrap(device_value.get());
  std::string summarized;
  TF_ASSERT_OK(unwrapped_handle->SummarizeValue(summarized));
  EXPECT_THAT(summarized, HasSubstr("\"CPU:0\": 3"));
}

}  // namespace parallel_device

But all this functionality of using asynchronous code with `async` and `await` is many times summarized as using "coroutines". It is comparable to the main key feature of Go, the "Goroutines".

## Conclusion

Let's see the same phrase from above:

    to control the order of logical devices (GPU only).

*   `tf.keras`:

    *   Changed the TensorBoard tag names produced by the
        `tf.keras.callbacks.TensorBoard` callback, so that summaries logged
        automatically for model weights now include either a `/histogram` or
        `/image` suffix in their tag names, in order to prevent tag name
        collisions across summary types.

* Support struct,array,slice types when sorting kubectl output ([#25022](https://github.com/kubernetes/kubernetes/pull/25022), [@zhouhaibing089](https://github.com/zhouhaibing089))
* federated api servers: Adding a discovery summarizer server ([#20358](https://github.com/kubernetes/kubernetes/pull/20358), [@nikhiljindal](https://github.com/nikhiljindal))

* Make discovery summarizer call servers in parallel ([#26705](https://github.com/kubernetes/kubernetes/pull/26705), [@nebril](https://github.com/nebril))

- Kube-scheduler now logs plugin scoring summaries at --v=4 ([#99411](https://github.com/kubernetes/kubernetes/pull/99411), [@damemi](https://github.com/damemi)) [SIG Scheduling]