Sentence Transformers Adds Support for Late Interaction Models

ariG23498 · x · 2026-08-18

Sentence Transformers v6.0 introduces MultiVectorEncoder, supporting ColBERT-style late interaction retrieval. Unlike standard embedding models that compress text into a single vector, multi-vector models keep one vector per token and score queries against documents using the MaxSim operator. This preserves token-level matching information, typically yielding stronger retrieval at the cost of larger indexes. It also enables state-of-the-art visual document retrieval by matching text queries directly to page images without OCR. The blog post provides code snippets for loading checkpoints, encoding, scoring, and integrating into a search stack.

Related event: Sentence Transformers v6.0 Brings Late-Interaction Multivector Retrieval(28 posts)→

Original post →

More from Apps

Apps channel →