rerank-2.5 and rerank-2.5-lite: instruction-following rerankers
12 days ago
- #AI
- #Machine Learning
- #Search Technology
- Introduction of rerank-2.5 series with improved performance and instruction-following capabilities.
- Significant accuracy improvements: rerank-2.5 and rerank-2.5-lite outperform Cohere Rerank v3.5 by 7.94% and 7.16% respectively.
- New 32K token context length support, 8x more than Cohere Rerank v3.5.
- Instruction-following feature allows natural language steering of relevance scores.
- Performance gains demonstrated on MAIR benchmark and in-house datasets.
- Detailed evaluation across multiple domains and languages showing consistent improvements.
- Examples provided showing practical applications of instruction-following in search queries.