dac-research/extra_evals_v1

extra_evals_v1 是一个综合评估数据集，结合了 LongBench v2、LooGLE、RULER 和 ZeroScrolls 等多个基准测试，用于长文档评估。该数据集包含多个任务子集，每个子集都有特定的配置和测试分割。数据集结构详细描述了各子集的列信息、令牌长度统计信息以及是否适合使用 LLM 评判的可行性。该数据集旨在评估各种领域和任务类型的长文档任务。

extra_evals_v1 is a curated evaluation dataset combining multiple benchmarks (LongBench v2, LooGLE, RULER, and ZeroScrolls) for long-document evaluation. It includes various task subsets, each with specific configurations and test splits. The dataset structure details the columns, token length statistics, and LLM-judge feasibility for each subset. It is designed to evaluate long-document tasks across diverse domains and task types.