dac-research/longbench_synthetic_v3

LongBench Synthetic V3是一个包含多个子集的数据集，用于评估和训练自然语言处理模型。数据集包含2wikimqa、gov_report、hotpotqa等多个子集，每个子集有不同的数据文件和分割（train、validation、test）。统计数据包括唯一上下文数量、样本行数、不同长度区间的样本数量以及标记统计信息。

LongBench Synthetic V3 is a dataset containing multiple subsets for evaluating and training natural language processing models. The dataset includes subsets such as 2wikimqa, gov_report, hotpotqa, etc., each with different data files and splits (train, validation, test). Statistics include the number of unique contexts, sample rows, sample counts in different length intervals, and token statistics.