on Ampere based GPUs. TensorFloat-32, or TF32 for short, is a math mode for
    NVIDIA Ampere based GPUs and is enabled by default.

*   A major refactoring of the internals of the Keras Functional API has been
    completed, that should improve the reliability, stability, and performance
    of constructing Functional models.

*   Keras mixed precision API

work, being done in conjunction with SIG Cloud Provider, is focusing on moving internal APIs that the OpenStack (and other providers) depends upon to staging to guarantee API stability. This work also included abstracting the in-tree Cinder API and refactoring code to the external Cinder provider to remove additional Cinder volume provider code.

Additional work was also done to implement an OpenStack driver for the Cluster API effort lead by SIG Cluster Lifecycle. For the external Cloud-Provider-OpenStack...

* Refactoring reorganize taints function in kubectl to expose operations ([#43171](https://github.com/kubernetes/kubernetes/pull/43171), [@ravisantoshgudimetla](https://github.com/ravisantoshgudimetla))

* vSphere cloud provider: vSphere cloud provider code refactoring ([#49164](https://github.com/kubernetes/kubernetes/pull/49164), [@BaluDontu](https://github.com/BaluDontu))