protologue

ColBERT

Also called late interaction, late-interaction retrieval, multi-vector retrieval.

ColBERT is a retrieval model that embeds every token of the query and of each passage separately and scores relevance by matching each query token to its most similar passage token, an approach called late interaction.

Description

Late interaction sits between single-vector dense retrieval, which is fast but coarse, and cross-encoder reranking, which is accurate but slow. Passage token embeddings can be precomputed and indexed, so ColBERT keeps much of a cross-encoder's precision at search-engine speed, at the cost of larger indexes.

Sources

  1. Khattab and Zaharia (2020). ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT.

Cite this entry

Protologue. (2026). ColBERT. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0130). https://protologue.com/t/colbert/

BibTeX
@misc{protologue_colbert,
  title = {ColBERT},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0130},
  url = {https://protologue.com/t/colbert/}
}

Markdown JSON