Trelis/touch-rugby-benchmark

该数据集包含了多个配置的数据，每个配置都有不同的特征和用途：
1. chunked配置：包含文档ID、文档文本、文档文件名、文档元数据（文件大小）、原始文档摘要、文档摘要、摘要模型、文本块信息、多跳文本块信息、块信息指标和块处理模型等特征。
2. ingested配置：包含文档ID、文档文本、文档文件名和文档元数据（文件大小）等特征。
3. lighteval配置：包含问题、真实答案、问题类别、问题类型、估计难度、引用、文档ID、文本块ID、问题生成模型、文本块和文档文本等特征。
4. multi_hop_questions配置：包含文档ID、源文本块ID、问题、自我答案、估计难度、自我评估问题类型、生成模型、思考过程和引用等特征。
5. single_shot_questions配置：包含文本块ID、文档ID、问题、自我答案、估计难度、自我评估问题类型、生成模型、思考过程和原始响应等特征。
6. summarized配置：包含文档ID、文档文本、文档文件名、文档元数据（文件大小）、原始文档摘要、文档摘要和摘要模型等特征。每个配置都有训练集分割，且数据集提供了不同大小的下载和实际数据集大小信息。

The dataset contains multiple configurations, each with different features and purposes:
1. chunked configuration: includes features like document ID, document text, document filename, document metadata (file size), raw document summary, document summary, summarization model, text chunk information, multi-hop text chunk information, chunk information metrics, and chunking model.
2. ingested configuration: includes features like document ID, document text, document filename, and document metadata (file size).
3. lighteval configuration: includes features like question, ground truth answer, question category, question type, estimated difficulty, citations, document ID, chunk IDs, question generating model, chunks, and document text.
4. multi_hop_questions configuration: includes features like document ID, source chunk IDs, question, self answer, estimated difficulty, self-assessed question type, generating model, thought process, and citations.
5. single_shot_questions configuration: includes features like chunk ID, document ID, question, self answer, estimated difficulty, self-assessed question type, generating model, thought process, and raw response.
6. summarized configuration: includes features like document ID, document text, document filename, document metadata (file size), raw document summary, document summary, and summarization model. Each configuration has a training set split, and the dataset provides different download sizes and actual dataset size information.