Model that scores a query and a passage together in one forward pass.
Для этих двух технологий пока не зафиксированы сравнимые характеристики.
Сформировано из различий выше по фиксированным правилам, а не моделью.
Choose Cross-Encoder if…
Ни одна из зафиксированных характеристик не говорит в его пользу.
Choose Bi-Encoder if…
Ни одна из зафиксированных характеристик не говорит в его пользу.
Страница остаётся доступной, но исключена из поисковых индексов, пока не наберёт достаточно подробностей со ссылками на источники (PRD SEO-08).
concept · Techniques
A cross-encoder concatenates the query and one candidate passage and runs them through the model together, producing a relevance score for that pair. It solves the problem that independent embeddings cannot represent how a specific query relates to a specific passage. It exists as the accurate but expensive half of retrieval: accurate enough to reorder a shortlist, too slow to search a corpus with.
concept · Techniques
A bi-encoder encodes the query and each passage independently into vectors, and relevance is their similarity. It solves the problem that a cross-encoder must run once per candidate, which is impossible against a large collection. It exists as the half of retrieval that can be precomputed and indexed, at the cost of losing the interaction between query and passage.
