Turbovec – Google's TurboQuant for vector search in Rust
Article URL: https://github.com/RyanCodrai/turbovec Comments URL: https://news.ycombinator.com/item?id=49349898 Points: 289 # Comments: 34
Google's Turbovec is a vector search index for Rust, akin to Google's TurboQuant for vector search. turbovec operates on 2-D float32 arrays, with vectors and queries presented as such, and other data types rejected in favor of explicit casting. turbovec is designed to fit large document corpora into memory efficiently, offering faster search performance compared to FAISS.
It utilizes Google Research's TurboQuant algorithm, a data-oblivious quantizer with minimal distortion and no separate training phase. Notably, turbovec supports stable ids that survive deletes using IdMapIndex, while filtering results through candidate sets. Filtering occurs within a SIMD kernel at 32-vector block granularity, with non-allowed slots dropped to minimize SIMD cost. turbovec offers drop-in replacements for in-tree reference vector/document stores, preserving the same public surface, persistence semantics, and retriever/pipeline wiring as the original stores.
The index supports stable external ids that survive deletes, with TurboQuant outperforming FAISS IndexPQ in various scenarios. The results demonstrate TurboQuant's efficiency and effectiveness in search tasks, offering comparable or superior performance over FAISS in specific use cases.
Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.