DS-Serve: A framework for efficient, scalable neural retrieval (berkeley-large-rag.github.io)