Jina Reranker v2 Base - Multilingual

Jina Reranker v2 model is a neural text reranking model, designed to enhance the relevance of search results.It complements text embedding models and refines search results by prioritizing documents relevant to a query.
This state-of-the-art reranker model enables a variety of applications that rely on precise search results across different languages, improved information retrieval, and high document throughput.
Use-cases: Vector search, retrieval augmented generation.

Hightlights:

High multilingual performance across the board: This reranker model ranks at the top compared to its competitors, in terms of 'Mean Reciprocal Rank' (MRR), recall and NDCG, according to BEIR, MKQA and AirBench.

Agentic RAG: This reranker model supports function-calling and text-to-SQL aware document reranking for agentic RAG.

Ultra-fast: 15x more documents throughput than bge-reranker-v2-m3, and 6x more than jina-reranker-v1-base-en.

Usage:

Please refer to this link for detailed usage.