Reproduce “GINGER: Spicing Up Graph-Based Vector Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daniel Salwasser, Lars Gottesbüren, Laxman Dhulipala, Jakub Łącki, Guy Blelloch, Rajesh Jayaram, Tobias Rubel, Kshiteej Sheth and 2 more
Ginger retrieves nearest neighbors from large embedding collections using graph search. Its Nested TurboQuant code uses a 4-bit prefix to navigate and nested refinement bits to rerank candidates, while losslessly compressing graph links. At 95% 10@10 recall (the fraction of true top ten recovered), Ginger averages 1.8× fastest-competitor throughput, leading on 13 of 14 benchmarks; SVS is 4% faster on BigANN.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.