Ray Serve LLM

Scalable LLM serving on Ray clusters with continuous batching, streaming, and multi-model routing.