Ray Serve

Scalable model-serving library built on Ray. Composes multiple ML models and Python functions into a single production endpoint with fractional GPU support, batching, model multiplexing, and autoscaling. Framework-agnostic — works with any Python ML library.