Architecture
Bi-Encoder
Architecture that encodes query and document independently â fast for retrieval at scale.
Definition
A bi-encoder (a.k.a. dual encoder) embeds queries and documents separately, then compares them via cosine similarity. This lets the corpus be indexed offline, enabling sub-millisecond retrieval at the cost of some accuracy versus a cross-encoder.
Common use cases
- Vector search
- RAG retrieval
- Recommendations