IVUL-KAUST/AutoThink

AutoThink数据集是一个包含数学推理、图像推理和视频推理的集合，例如DAPO-MATH、ViRL和Video-R1。该数据集通过从TVBench、MMR-Vbench和NExT-GQA中筛选出高质量推理样本，并排除了VideoVista、ShortVid-Bench和Video-Holmes数据集，因为这些数据集中含有大量低质量推理样本，不适合用于AutoThink。这个子集是为了满足AutoThink的需求而创建的。The AutoThink dataset is a collection of math-reasoning, image-reasoning, and video-reasoning datasets, including DAPO-MATH, ViRL, and Video-R1. The dataset was created by filtering high-quality reasoning samples from TVBench, MMR-Vbench, and NExT-GQA, and excluding VideoVista, ShortVid-Bench, and Video-Holmes due to the presence of numerous low-quality reasoning samples that are not suitable for AutoThink. This subset was created to meet the needs of AutoThink.