Expand description
Variable-length multi-vector document storage with Meta Token mode.
MultiVectorStore holds a collection of MultiVectorDoc entries where
each document carries one or more embedding vectors. In MetaToken mode
every document has exactly k vectors (MetaEmbed learnable summary tokens);
in PerToken mode the count is unconstrained (naive ColBERT).
Structs§
- Multi
Vector Doc - A document represented by one or more embedding vectors.
- Multi
Vector Store - In-memory store for multi-vector documents.
Enums§
- Multi
VecMode - Controls how many vectors a document may contain and what they represent.
- Multivec
Error - Errors produced by
MultiVectorStore.