PGPU 2.0.0

• use cuVS for kmeans
• remove FAISS entirely
• implement multi-level clustering
  • supporting 1 or 2 levels
• re-engineer the batch execution
  • ensure batches are aligned with kmeans runs
  • calculate number of intermediate results needed
• BREAKING: change interfaces to support multi-level clustering

What's Changed

• Use cuVS for kmeans clustering by @timwaizenegger in #3
• optimize batch output cluster count by @timwaizenegger in #4
• better readme by @timwaizenegger in #5
• Multi level index by @timwaizenegger in #6
• update cuvs from 25.10.00 to 25.12.00 by @felipe-mendes-edb in #7
• prepare version 2.0.0 by @timwaizenegger in #8

New Contributors

• @felipe-mendes-edb made their first contribution in #7

Full Changelog: v1.0.0...v2.0.0

v1.0.0

The initial release of PGPU 🎉 🚀

• vector index build acceleration for vchord; building centroids on the GPU for "external build" on vchord
• uses FAISS for GPU acceleration
• PG native table scan to read training vectors
• all running in a single SQL function call