vLLM patches critical CVE‑2026‑0994 in urgent v0.19.1 release

- vLLM published v0.19.1, a critical release that patches CVE‑2026‑0994 (protobuf deserialization) and adds async scheduling, Gemma 4 support and KV cache optimisations. - The update also notes Kubernetes‑focused improvements: integration with the llm‑d CNCF sandbox and compatibility with Karpenter v1.11.x for topology‑aware GPU scheduling. - The maintainers urged production upgrades for both security and distributed‑inference performance improvements. (x.com/kubesimplify/status/2047827934044672218) (x.com/i/status/2047571384247722008)