MAIR
A benchmark for instruction-following retrieval: 123 tasks, each with a task-specific instruction that the retriever or reranker must take into account (https://arxiv.org/abs/2410.10127).
Use in This Vault
rerank-3 and rerank-3-lite - the next generation of Voyage rerankers uses MAIR to check that instruction-following survived a model upgrade. Voyage rerank-3 scored on par with rerank-2.5 and 3.58% above Cohere Rerank v4.0 Pro.