RaBitQ-Library
An official lightweight library for the RaBitQ algorithm and its applications in vector search.
Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.
An official lightweight library for the RaBitQ algorithm and its applications in vector search.
Extended-RaBitQ is a software implementation of a novel quantization algorithm based on RaBitQ, designed for Approximate Nearest Neighbor Search in high-dimensional Euclidean space. This project was accepted at SIGMOD 2025 and has been archived in favor of the newly released RaBitQ-Library for further development. The algorithm compresses high-dimensional vectors with arbitrary compression rates ranging from 3 to 9 bits per dimension, including support for 4 and 8 bits. It achieves computation speed comparable to classical scalar quantization while delivering dominant accuracy, particularly excelling in the 2-bit to 6-bit range to enable high recall without the need for reranking. The source code supports AVX512 instructions for optimization and includes tools to create indexes and test search performance on datasets like openai1536 using inverted file indexes with a specified number of clusters. Users can generate indexes and evaluate query performance via command-line executables, with results stored in a d