ibm-research/rag-hpo-bench

该数据集名为RAG‑HPO Bench，包含了一篇关于检索增强生成（RAG）超参数优化方法分析的论文中的网格搜索结果。数据集涵盖了162种RAG配置在五个RAG QA数据集（AIArxiv、BioASQ、ClapNQ、MiniWiki和WatsonxQA）上的开发和测试分割的输出和评分。数据集还包括了不同配置的详细结果文件，探索的超参数空间（如块大小、块重叠、嵌入模型、Top-k和生成模型），以及使用的评估指标（如LLMaaJ-AC、Lexical-AC等）。此外，README还提供了数据集的引用信息、许可证（CC-BY-NC-SA 4.0）和变更日志。

This dataset, named RAG‑HPO Bench, contains the grid results from the paper "An Analysis of Hyper‑Parameter Optimization Methods for Retrieval Augmented Generation". It includes the per-configuration outputs and scores of 162 RAG configurations on development and held-out test splits across five RAG QA datasets (AIArxiv, BioASQ, ClapNQ, MiniWiki, and WatsonxQA). The dataset also includes detailed result files for each configuration, the explored hyperparameter space (e.g., chunk size, chunk overlap, embedding model, top-k, and generative model), and the evaluation metrics used (e.g., LLMaaJ-AC, Lexical-AC, etc.). Additionally, the README provides citation information, the license (CC-BY-NC-SA 4.0), and a changelog.