STEPS

STEPS是一个针对顺序任务中顺序推理的挑战性基准，由加州大学圣塔芭芭拉分校创建。该数据集包含约298,000条记录，主要来源于FOOD.COM的食谱数据。数据集的构建过程涉及对食谱的筛选和分类，确保数据集的质量和适用性。STEPS数据集主要用于评估大型语言模型在顺序任务中的顺序推理能力，特别是在确定下一步行动的合理性方面。

STEPS is a challenging benchmark for sequential reasoning in sequential tasks, developed by the University of California, Santa Barbara. This dataset contains approximately 298,000 records, primarily sourced from recipe data on FOOD.COM. The construction of the STEPS dataset involves filtering and categorizing recipes to ensure its quality and applicability. The STEPS dataset is primarily designed to evaluate the sequential reasoning capabilities of Large Language Models (LLMs) in sequential tasks, particularly in determining the plausibility of the proposed next step.