### Bug or Regression

- Bump node-problem-detector version to v0.8.5 to fix OOM detection in with Linux kernels 5.1+ ([#96716](https://github.com/kubernetes/kubernetes/pull/96716), [@tosi3k](https://github.com/tosi3k)) [SIG Cloud Provider, Scalability and Testing]

## Breaking Changes

*   `tf.keras`:
    *   In `tf.keras.applications` the name of the "top" layer has been
        standardized to "predictions". This is only a problem if your code
        relies on the exact name of the layer.
    *   Huber loss function has been updated to be consistent with other Keras
        losses. It now computes mean over the last axis of per-sample losses

    * This can solve nuanced errors when using middlewares. Before Starlette `0.24.0`, a new instance of each middleware class would be created when a new middleware was added. That normally was not a problem, unless the middleware class expected to be created only once, with only one instance, that happened in some cases. This upgrade would solve those cases (thanks [@adriangb](https://github.com/adriangb)! Starlette PR [#2017](https://github.com/enc...

TestModulusAndNatSiz(t *testing.T) { // These are 126 bit (2 * _W on 64-bit architectures) values, serialized as // 128 bits worth of bytes. If leading zeroes are stripped, they fit in two // limbs, if they are not, they fit in three. This can be a problem because // modulus strips leading zeroes and nat does not. m := modulusFromBytes([]byte{ 0x3f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}) xb := []byte{0x3f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,...