KServe provider introduces multi-node LLM head and worker deployments

The KServe provider now supports multi-node LLM deployments with dedicated head and worker nodes, enabling inference on models that exceed the memory capacity of a single node.

Previously, LLM deployments were constrained to a single node, which limited the maximum model size to available GPU memory on one machine. For large models, this meant either choosing a smaller model or provisioning prohibitively expensive high-memory nodes.

Multi-node LLM support is available now in the KServe provider. To learn more, see the release notes.