* Match GroupVersionKind against specific version ([#34010](https://github.com/kubernetes/kubernetes/pull/34010), [@soltysh](https://github.com/soltysh))
* fix yaml decode issue ([#34297](https://github.com/kubernetes/kubernetes/pull/34297), [@AdoHe](https://github.com/AdoHe))

*   Added a distributed runtime using GRPC
*   Move skflow to `contrib/learn`
*   Better linear optimizer in `contrib/linear_optimizer`
*   Random forest implementation in `contrib/tensor_forest`
*   CTC loss and decoders in `contrib/ctc`
*   Basic support for `half` data type
*   Better support for loading user ops (see examples in `contrib/`)
*   Allow use of (non-blocking) Eigen threadpool with

([#64248](https://github.com/kubernetes/kubernetes/pull/64248), [@feiskyer](https://github.com/feiskyer))

### SIG CLI

* You can now use the `base64decode` function in kubectl go templates to decode base64-encoded data, such as `kubectl get secret SECRET -o go-template='{{ .data.KEY | base64decode }}'`. ([#60755](https://github.com/kubernetes/kubernetes/pull/60755), [@glb](https://github.com/glb))

- `kubeadm`: improved the logic that checks whether a `systemd` service exists. ([#120514](https://github.com/kubernetes/kubernetes/pull/120514), [@fengxsong](https://github.com/fengxsong))
- `kubeadm`: will now use universal deserializer to decode static pod. ([#120549](https://github.com/kubernetes/kubernetes/pull/120549), [@pacoxu](https://github.com/pacoxu))
- `kubectl prune v2`: Switched annotation from `contains-group-resources` to `contains-group-kinds`,

  request. A resource is considered corrupt if it can not be successfully retrieved from the storage due to
  a) transformation error e.g. decryption failure, or b) the object failed to decode. Normal deletion flow is
  attempted first, and if it fails with a corrupt resource error then it triggers unsafe delete.
  In addition, when this feature is enabled, the 'details' field of 'Status' from the LIST response

- The `apiserver.latency.k8s.io/decode-response-object` annotation was added to the audit log to record the decoding time. ([#121512](https://github.com/kubernetes/kubernetes/pull/121512), [@HirazawaUi](https://github.com/HirazawaUi))

- Add a helper function to decode scheduler plugin args. ([#80696](https://github.com/kubernetes/kubernetes/pull/80696), [@hex108](https://github.com/hex108))

* Allows changing nodeName in endpoint update. ([#68575](https://github.com/kubernetes/kubernetes/pull/68575), [@prameshj](https://github.com/prameshj))
* kube-apiserver would return 400 Bad Request when it couldn't decode a json patch. ([#68346](https://github.com/kubernetes/kubernetes/pull/68346), [@CaoShuFeng](https://github.com/CaoShuFeng))
    * kube-apiserver would return 422 Unprocessable Entity when a json patch couldn't be applied to one object.

~[n]fieldElement](b []byte, f T) []byte { out, B := sliceForAppend(b, encodingSize12) for i := 0; i < n; i += 2 { x := uint32(f[i]) | uint32(f[i+1])<<12 B[0] = uint8(x) B[1] = uint8(x >> 8) B[2] = uint8(x >> 16) B = B[3:] } return out } // polyByteDecode decodes the 384-byte encoding of a polynomial, checking that // all the coefficients are properly reduced. This fulfills the "Modulus check" // step of ML-KEM Encapsulation. // // It implements ByteDecode₁₂, according to FIPS 203, Algorithm 6. func polyByteDecode[T...

* Protobuf serialization does not distinguish between `[]` and `null`.