High-performance KNN library pynear 2.3 released

·2026.05.23 20:27

Key point

pynear 2.3, a high-performance SIMD-accelerated KNN library compatible with scikit-learn, has been released.

pynear is a fast KNN (K-Nearest Neighbors) library that can be used without trading off accuracy for production speed.

Key features are as follows:

Drop-in replacement compatibility with scikit-learn's metric='cosine' option
Performance optimization through SIMD acceleration
For binary descriptors, delivers up to 257x faster speed than Faiss at 100% recall
Support for cosine indexes (exact and approximate methods)
Includes IndexBinaryMultiHash, a fast approximate index with high recall

It can be installed with the pip install -U pynear command.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.