179 questions
No questions match those filters.
← All questions
Retrieval & EmbeddingsMedium· ColBERT· Late Interaction· Information Retrieval
What is 'late' about late interaction in models like Co...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe ‘lateness’ refers to the invariant that the document representation is query-independent, allowing it to be precomputed. Unlike cross-encoders, ColBERT retains per-token vectors on both sides, which are then compared using a MaxSim operation. This architecture is not merely a ‘cheaper cross-encoder’ but a distinct approach that restricts interaction to a bilinear form over precomputed vectors.