Multi-Vector Embeddings

An information retrieval approach using dense vector representations (embeddings) to encode queries and documents. Unlike sparse methods like BM25, dense retrieval captures semantic meaning in continuous vector spaces, enabling neural search and forming the foundation of modern RAG systems.

A retrieval paradigm where query and document encodings are kept separate until a late interaction stage, enabling more expressive and efficient similarity computations. Pioneered by ColBERT and extended by ColPali and ColQwen, this approach maintains fine-grained representations while enabling fast retrieval.

A search paradigm where queries and documents are encoded differently, optimized for scenarios where queries are short and documents are long. Common in information retrieval and modern embedding models designed specifically for search.

Retrieval paradigm where query and document tokens are encoded separately and interactions computed at search time, combining efficiency of bi-encoders with expressiveness of cross-encoders.

Python library designed to simplify the integration and training of state-of-the-art late-interaction retrieval methods, particularly ColBERT, within RAG pipelines with a modular and user-friendly interface.