RAG Reranking Benchmarks
收藏Zenodo2026-06-12 更新2026-06-17 收录
下载链接：
https://zenodo.org/doi/10.5281/zenodo.20669912
下载链接
链接失效反馈官方服务：
资源简介：
480 timing measurements across 4 reranking model families (Cohere Rerank 3.5, BGE-Reranker-v2-m3, ms-marco-MiniLM-L-12-v2, cross-encoder/nli-deberta-v3-small) and 2 providers benchmarking reranking latency and retrieval accuracy in RAG pipelines. Data include per-query latency samples (CSV), model configuration metadata, ANOVA statistical outputs, and 32 scored retrieval queries across 4 categories (n=8 each) with 98.1% ground truth accuracy. Reranking adds a mean 31ms overhead (65% increase in retrieval latency) while delivering 6-8% accuracy improvement over single-method search; full statistical methodology and query scoring rubric are included. Supplementary dataset for the blog post: Clouatre, H. (2026). RAG Reranking Benchmarks: Supplementary Materials.
提供机构：
Zenodo创建时间：
2026-06-12
