$(this._dialog)\n        .one(Util.TRANSITION_END, transitionComplete)\n        .emulateTransitionEnd(transitionDuration)\n    } else {\n      transitionComplete()\n    }\n  }\n\n  _enforceFocus() {\n    $(document)\n      .off(Event.FOCUSIN) // Guard against infinite focus loop\n      .on(Event.FOCUSIN, (event) => {\n        if (document !== event.target &&\n            this._element !== event.target &&\n            $(this._element).has(event.target).length === 0) {\n          this._element.focus()\n...

- EndpointSlice has a new NodeName field guarded by the EndpointSliceNodeName feature gate.
  - EndpointSlice topology field will be deprecated in an upcoming release.
  - EndpointSlice "IP" address type is formally removed after being deprecated in Kubernetes 1.17.

* Remove Initializers from admission-control in kubernetes-master charm for pre-1.7 ([#46987](https://github.com/kubernetes/kubernetes/pull/46987), [@Cynerva](https://github.com/Cynerva))
* Added state guards to the idle_status messaging in the kubernetes-master charm to make deployment faster on initial deployment. ([#47183](https://github.com/kubernetes/kubernetes/pull/47183), [@chuckbutler](https://github.com/chuckbutler))

*   Add support of local soft device placement for eager op.
*   Add HW acceleration support for `LogSoftMax`.
*   Added a function `nested_value_rowids` for ragged tensors.
*   Add guard to avoid acceleration of L2 Normalization with input rank != 4
*   Add `tf.math.cumulative_logsumexp operation`.
*   Add `tf.ragged.stack`.
*   Fix memory allocation problem when calling `AddNewInputConstantTensor`.