Test of Time (ToT)
收藏资源简介:
Test of Time (ToT) 数据集由Google Research创建,专注于评估大型语言模型(LLMs)在时间推理任务中的表现。该数据集包含2800个问题,分为两个子任务:ToT-Semantic和ToT-Arithmetic。ToT-Semantic通过合成问题评估时间语义和逻辑理解,而ToT-Arithmetic则通过众包任务评估时间算术能力。数据集的创建过程涉及随机结构生成和问题生成,确保了问题的多样性和复杂性。ToT数据集的应用领域主要在于提升LLMs在时间推理方面的性能,解决现有数据集在时间推理评估上的局限性。
Test of Time (ToT) dataset was created by Google Research, focusing on evaluating the performance of Large Language Models (LLMs) on temporal reasoning tasks. The dataset contains 2,800 questions divided into two subtasks: ToT-Semantic and ToT-Arithmetic. ToT-Semantic evaluates temporal semantic and logical understanding through synthetic questions, while ToT-Arithmetic assesses temporal arithmetic capabilities via crowdsourced tasks. The dataset's creation process involves random structure generation and question generation, ensuring the diversity and complexity of the questions. The main application of the ToT dataset is to improve the temporal reasoning performance of LLMs and address the limitations of existing datasets in temporal reasoning evaluation.




