MATH-Perturb
收藏资源简介:
MATH-Perturb数据集是由普林斯顿大学和谷歌的研究人员构建的,包含279个经过简单扰动和困难扰动的数学问题,这些问题源自MATH数据集中难度最高的第五级别问题。数据集由12位具有强大数学背景的博士生注解,并通过严格的校验流程确保质量。该数据集旨在评估大型语言模型在面临困难扰动时的数学推理能力,对于推动未来语言模型鲁棒性和可靠性的发展具有重要意义。
The MATH-Perturb dataset was constructed by researchers from Princeton University and Google. It contains 279 math problems subjected to both simple and difficult perturbations, which are derived from the highest-difficulty Level 5 problems in the original MATH dataset. The dataset was annotated by 12 doctoral students with strong mathematical backgrounds, and its quality was guaranteed through a strict validation process. This dataset aims to evaluate the mathematical reasoning capabilities of large language models when confronted with challenging perturbations, and it is of great significance for advancing the development of robustness and reliability of future language models.




