DATETIME
收藏资源简介:
DATETIME数据集是一个高质量的数据集,旨在评估大型语言模型(LLM)在日期时间处理方面的翻译和推理能力。数据集包含日期和时间的组合,例如'11th.february.2023, 1:12:31'。虽然日期时间对于人类来说是直观和易于理解的,但对于LLM来说却具有挑战性。DATETIME数据集提供了翻译、计算和混合任务,以评估LLM在日期时间处理方面的能力。数据集涵盖了从1990年1月1日到9999年12月31日的日期时间范围,并具有多种自然表示形式和时区。数据集是合成生成的,以确保准确性和无噪声,并且可以频繁地重新生成以减少污染。DATETIME数据集的应用领域包括评估LLM在日期时间处理方面的能力,并促进相关研究的发展。
The DATETIME dataset is a high-quality corpus designed to evaluate the translation and reasoning capabilities of large language models (LLMs) in datetime processing. This dataset contains combinations of dates and times, such as '11th February 2023, 1:12:31'. While datetime is intuitive and easy for humans to comprehend, it poses considerable challenges for LLMs. The DATETIME dataset provides translation, calculation, and hybrid tasks to assess the datetime processing capabilities of LLMs. It covers a datetime range from January 1, 1990 to December 31, 9999, and includes various natural expression formats and time zones. This dataset is synthetically generated to ensure accuracy and noise-free quality, and can be frequently regenerated to mitigate data contamination. The application scenarios of the DATETIME dataset include evaluating the datetime processing capabilities of LLMs and promoting the development of relevant research.

- 1DATETIME: A new benchmark to measure LLM translation and reasoning capabilities瑞士苏黎世联邦理工学院(ETH Zurich) · 2025年



