LLMs & Generative AITop pick
LEANN Vector Index: How UC Berkeley Shrank a 201 GB RAG Index to 6 GB
LEANN, from UC Berkeley's Sky Computing Lab, keeps a pruned graph and recomputes embeddings only on the search path, shrinking a 60M-chunk RAG index from 201 GB to 6 GB. We explain how it works, check the math and test the idea with FAISS.
13 min read