PyNNDescent 0.7 benchmark: ~6x faster kNN graph builds on 64-core Xeon
leland_mcinnes · x · 2026-10-18
Leland McInnes (author of UMAP) teases major performance improvements coming to PyNNDescent, especially for large multicore jobs.
The teaser benchmarks show version 0.7 cutting kNN graph build times on a 64-core Xeon 2.8GHz from 131s to 19s on ArXiv Nomic (768d), 120s to 17s on ImageNet CLIP (512d), and 33s to 11s on SimpleWiki OpenAI (3072d), with similar gains on a Mac M2 12-core.
More from Research
- Interpretability debate: does scaling make models denser, or just bigger? — JacobHHilton · 2026-10-18
- Interp debate: denser wiring at scale is the real question, not 'more stuff' — banburismus_ · 2026-10-18
- Terence Tao: OpenAI's quasi-Riemann result is the biggest advance in 50 years — jracine22 · 2026-10-18
- LLMs compute addition via Fourier features: 28 LLaMA neurons implement the mechanism — banburismus_ · 2026-10-18
- What is wrong with AI writing - ICLR reviewing edition — bobrodsky · 2026-10-18
- Meta, MIT and Yale propose ReHarness: backward revision for agentic continual learning — HanGuo97 · 2026-10-18