lime-nlp/Synthetic_Unanswerable_Math

Synthetic Unanswerable Math (SUM)是一个高质量的隐式不可回答数学问题数据集，旨在探测和改进大型语言模型在遇到无法回答的问题时的拒绝行为。每个数据条目包括一个原始的可解答数学问题和一个经过合成的不可解答版本。这个数据集用于诊断、训练和教会模型在推理时计算其不确定性和知识边界，并在适当的时候放弃回答。

Synthetic Unanswerable Math (SUM) is a dataset of high-quality, implicitly unanswerable math problems constructed to probe and improve the refusal behavior of large language models (LLMs). Each entry in the dataset includes an original, solvable math problem and a synthetically modified version that is designed to be unsolvable. The dataset is used to diagnose, train, and teach models to reason about their own uncertainty and knowledge boundaries during inference and to abstain when appropriate.