Billion-scale ANNS Benchmarks

A benchmarking framework for evaluating approximate nearest neighbor search (ANNS) algorithms on billion-scale datasets. It is designed to assess the scalability and performance of vector databases and ANNS methods on very large datasets.

Features

• Provides a framework for benchmarking ANNS algorithms at the billion-scale level
• Supports evaluation of both algorithms and hardware for scalability and performance
• Includes datasets suitable for billion-scale benchmarking (details available on the project website)
• Resources and guides for running benchmarks and interpreting results
• Historical results and documentation for NeurIPS 2021 and NeurIPS 2023 competitions
• Tools for dataset preparation, evaluation, and result visualization
• Open-source and extensible for new datasets or algorithms

Category

• Benchmarks & Evaluation

Tags

benchmark, anns, scalability, performance

Source

https://github.com/harsha-simhadri/big-ann-benchmarks

Pricing

• No pricing information provided (open source project).