Bi-Encoder
Also called Dual Encoder
A model that embeds query and document separately, so you can precompute and search millions fast.
Think of it like
Two people writing independent summaries, then comparing notes — quick, but they never actually read together.
Example
A bi-encoder embeds every document once ahead of time; at query time you embed just the question and do a fast vector search.
How it actually works
A bi-encoder maps query and document to vectors independently, which is exactly what lets you embed the whole corpus in advance and retrieve with fast similarity search. The cost of that independence is missing fine-grained interactions between query and document, so it is less precise than a cross-encoder. The standard pattern pairs them: bi-encoder for fast recall, cross-encoder to rerank.
For product teams
The workhorse that makes large-scale semantic search fast; usually backed by a reranker for precision.
For engineers
Independent query/doc encoders enabling precomputed ANN; loses cross-attention precision, hence the rerank pairing.
Related
- Dense Retrieval — The basis of dense retrieval.
- Cross-Encoder — Traded off against the accurate cross-encoder.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome