Retrieval Augmentation and Semantic Search at Scale
Explore the intricacies of Vector Search and Retrieval-Augmented Generation in this 32-minute talk by Ash Vardanian from Unum Cloud. Delve into technical benchmarks and uncover the challenges that hinder various solutions from scaling effectively. Gain insights from multiple CLIP-like AI pre-training experiments and learn about serving over 10 Billion vectors from a single machine. Examine the design decisions behind the USearch and UForm open-source libraries, and discover answers to crucial questions such as selecting the optimal GPU for inference workloads and serving search results from SSDs instead of RAM. Understand the tools and techniques that enable efficient retrieval augmentation and semantic search at scale, addressing the high costs associated with AI work.