* `tf.lite`
    * LiteRT announced a [new release](https://developers.googleblog.com/en/litert-maximum-performance-simplified/) at Google IO ‘25 that improves upon TFLite, particularly in terms of NPU and GPU hardware acceleration and performance for on-device ML and AI applications. The APIs are available in Kotlin and C++.

- AcceleratorStats will be available in the Summary API of kubelet when cri_stats_provider is used. ([#97017](https://github.com/kubernetes/kubernetes/pull/97017), [@ruiwen-zhao](https://github.com/ruiwen-zhao)) [SIG Node]
- Exposes and sets a default timeout for the SubjectAccessReview client for DelegatingAuthorizationOptions ([#95910](https://github.com/kubernetes/kubernetes/pull/95910), [@p0lyn0mial](https://github.com/p0lyn0mial)) [SIG API Machinery and Cloud Provider]

* 📝 Update docs for testing, fix examples with relative imports. PR [#5302](https://github.com/tiangolo/fastapi/pull/5302) by [@tiangolo](https://github.com/tiangolo).

### Translations

errors.New("crypto/rsa: invalid prime") } if pN.Mul(qN, N).IsZero() != 1 { return errors.New("crypto/rsa: p * q != n") } // Check that de ≡ 1 mod p-1, and de ≡ 1 mod q-1. // // This implies that e is coprime to each p-1 as e has a multiplicative // inverse. Therefore e is coprime to lcm(p-1,q-1) = λ(N). // It also implies that a^de ≡ a mod p as a^(p-1) ≡ 1 mod p. Thus a^de ≡ a // mod n for all a coprime to n, as required. // // This checks dP, dQ, and e. We don't check d because it is not actually //...