GEOTEMP
收藏资源简介:
GEOTEMP是一个包含32万个提示的数据集,覆盖了217个国家中的289个城市和37个时区,旨在评估语言模型在时空推理方面的能力。数据集包含了四种任务模板,用于测试不同水平的时空推理。数据集的创建过程包括收集时区和城市信息、设计任务模板以及生成测试集。数据集可用于评估语言模型在处理时间和地点问题时的性能,以及探索如何通过注入时空知识来提高模型性能。
GEOTEMP is a dataset containing 320,000 prompts, covering 289 cities across 217 countries and 37 time zones, designed to evaluate the spatial-temporal reasoning capabilities of language models. The dataset includes four task templates to test spatial-temporal reasoning at different levels. Its creation process involves collecting time zone and city information, designing task templates, and generating the test set. This dataset can be used to evaluate the performance of language models when handling temporal and spatial questions, as well as to explore how to improve model performance by injecting spatial-temporal knowledge.




