What Are Embeddings and Rerankers?
Embeddings
Section titled “Embeddings”Embeddings convert text into numerical vectors so semantic similarity becomes measurable. A typical use case is search over a vector database.
noris model: Harrier OSS v1 0.6B
Rerankers
Section titled “Rerankers”A reranker is a second-stage filter: it scores query and passage pairwise, producing a more precise ranking than embedding search alone.
noris models:
- BGE Reranker v2 M3 (XLM-RoBERTa, 0.6B)
- Jina Reranker v2 (0.3B)
Why two models? They’re built on different architectures and profiles, covering different language and domain requirements.
Embeddings and rerankers are often the invisible part of a RAG solution: without good retrieval, even the best LLM produces weak answers. It pays to invest in these two components, they largely determine the quality of your results.