Module multi_vector 

Multi-vector (ColBERT) index support.

Builds on the existing HNSW backend by flattening all token vectors from all documents into a single index (one HNSW node per token), then aggregating results per-document at query time using the ColBERT MaxSim formula:

score(Q, D) = Σ_i max_j (q_i · d_j)

Structs

MultiVectorBuilder

Builds a multi-vector index from per-document token embeddings.

MultiVectorResult

A multi-vector search result (one per document).

MultiVectorSearcher

A loaded multi-vector index, ready for MaxSim search.

TokenLabel

Metadata for a single token vector in the flattened index.