VectorDB-NTU/RaBitQ-Library

RaBitQ implements a vector quantization algorithm achieving optimal theoretical error bounds with ultra-low bit-widths (1-7 bits per dimension), supporting fast distance estimation via bitwise operations. It integrates with popular vector search indexes—IVF, HNSW, and QG—enabling memory-efficient approximate nearest neighbor search with 90-99% recall depending on bit configuration. The library provides CPU implementations in C++ and bindings across multiple production systems including Milvus, Faiss, VSAG, and Elasticsearch.

129 stars.