Expand description
WARP index with IVF structure
This module implements the WARP index which organizes compressed token embeddings by centroid for cache-efficient search. The index supports:
- Training from sample embeddings
- Incremental insertion of documents
- Building (compacting) for efficient search
- MaxSim-based multi-vector search
Structsยง
- Warp
Index - WARP index for efficient multi-vector retrieval.