TRAC

TRAC数据集是由中山大学计算机科学系创建，专注于动作和变化推理的文本基准。该数据集包含15000个样本，涵盖四种基本的推理任务：预测动作效果、判断动作适用性、制定有效计划和识别目标。数据集通过控制复杂度，专注于动作和变化的推理，旨在评估语言模型在结构泛化能力上的表现。TRAC数据集的应用领域主要集中在人工智能中对动态环境理解和交互的关键问题，特别是动作和变化的基本知识处理。

The TRAC dataset, a text benchmark dedicated to reasoning about actions and changes, was developed by the Department of Computer Science at Sun Yat-sen University. It contains 15,000 samples covering four fundamental reasoning tasks: predicting the effects of actions, judging the applicability of actions, formulating effective plans, and identifying goals. By controlling for complexity, the dataset focuses on reasoning about actions and changes, aiming to evaluate the structural generalization capabilities of language models. The application scenarios of the TRAC dataset primarily focus on key issues in artificial intelligence related to dynamic environment understanding and interaction, particularly the processing of basic knowledge about actions and changes.