Turbovec – Google's TurboQuant for vector search in Rust

Turbovec is a Rust vector index with Python bindings, built on Google Research's TurboQuant algorithm. It is designed for data-oblivious quantization with near-optimal distortion and no separate training phase. turbovec can handle a 10 million document corpus using only 4 GB of RAM in float32 format, compared to 31 GB required by FAISS.

It allows for stable ids that survive deletes using IdMapIndex, and enables filtering results to a candidate set produced by another system. turbovec provides drop-in replacements for in-tree reference vector/document stores in each framework and achieves faster search times than FAISS, including for larger vector dimensions. On ARM architecture, TurboQuant beats FAISS FastScan in every configuration, averaging 3.5 times faster at 4-bit and 26% faster at 2-bit.

On x86 architecture, TurboQuant also outperforms FAISS in every configuration, averaging 3.4 times faster at 4-bit and 20% faster at 2-bit. The insertion and removal latencies of turbovec are significantly faster than those of FAISS, with turbovec consuming less time per operation. Turbovec serializes to a single .tv file with an fsync + atomic rename, while FAISS uses a write_index / read_index on the precision-matched IndexPQFastScan.