Did UC Berkeley release LEANN, an open-source vector index that shrinks a 60M-chunk RAG index from about 201 GB to 6 GB while matching HNSW accuracy?
UnconfirmedLast checked
UC BerkeleyFirst seen
Verdict
MarkTechPost summarizes the LEANN paper/repo and MLSys 2026 Best Research Paper award; primary paper numbers not re-run here.
Evidence timeline · 1 source · newest first
Unconfirmed
MarkTechPost covers UC Berkeley Sky Computing Lab’s LEANN, an MIT-licensed vector index that stores a pruned graph and recomputes embeddings at query time, reporting a 60M-chunk Wikipedia index cut from 201 GB to 6 GB while matching HNSW accuracy at 90% recall; the paper won Best Research Paper at MLSys 2026.