AnalysisDevelopersOctober 27, 2026

LEANN vector index shrinks 60M-chunk RAG index from 201 GB to 6 GB

Read original source →marktechpost.com

UC Berkeley's Sky Computing Lab built LEANN, which recomputes embeddings at query time instead of storing them, cutting a 60M-chunk Wikipedia index to 6 GB. It matches HNSW accuracy at 90% recall using under 5% of raw-data storage; the paper won Best Research Paper at MLSys 2026.

1 source

More stories today

Open the live feed