ulab-ai/Time-Bench

Time-Bench数据集是一个针对时序推理任务的数据集，用于训练和评估语言模型在时序理解、未来事件时间预测以及创造性未来场景生成方面的能力。数据集包含了多个阶段所需的文件，用于不同的训练和验证目的，包括简单时序推理任务、时间戳推断、时间差估计、事件排序、遮蔽时间实体补全、未来事件时间预测以及与真实世界新闻文章的比较分析。

The Time-Bench dataset is designed for temporal reasoning tasks, training, and evaluating language models in their ability to understand temporal relations, predict future event times, and generate creative future scenarios. The dataset consists of multiple files for different stages of training and validation purposes, including simple temporal inference tasks, timestamp inference, time-difference estimation, event ordering, masked time entity completion, future event time prediction, and comparative analysis with real-world news articles.