Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Source
huggingface.co
Date
Why it matters
Late interaction preserves tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →-level matching that a single dense vector averages away — stronger retrieval on rare entities and exact identifiers, and the state of the art for searching page images without OCR, at the cost of a larger index.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
articleTraining and Finetuning Multi-Vector Embedding Models with Sentence Transformershuggingface.co
articleDenseOn with the LateOn: Fully Open Dense and Late-Interaction Models for Multilingual, Long-Context, and Code SearchRapha\"el Sourty, Antoine Chaffin, Paulo Roberto Moura Junior, Am\'elie Chatelain
postPerplexity open-sources pplx-embed-v2-late multimodal retrieval modelsPerplexity