Home

Super K-Means

SuperKMeans is a super-fast library for clustering high-dimensional vector embeddings. The main use case is to create partition-based indexes in large vector collections (e.g., IVF).

We offer two clustering options:

• Vanilla Super K-Means: Classical Lloyd's k-means accelerated with dimension pruning. Clustering is of equivalent quality to FAISS.
  • Usage example in C++
  • Usage example in Python
  • API Documentation
• Hierarchical Super K-Means: Hierarchical k-means accelerated with dimension pruning. Extremely fast for large collections (+1M embeddings) with minimal quality loss.
  • Usage example in C++
  • Usage example in Python
  • API Documentation

Use Super K-Means if:

• You need to index a large collection of high-dimensional (d > 128) vector embeddings
• You need a lightweight and much faster alternative to FAISS clustering

How to Install

Check INSTALL.md.

Documentation (C++ API)

• Vanilla SuperKMeans
• Hierarchical SuperKMeans