10K multimodal math dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Osilly/Vision-R1
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了大约1万个多模态数学数据，这些数据被用于强化学习训练Vision-R1模型。在GRPO训练过程中，这些数据也被利用来提升模型的推理能力。数据规模为1万条，任务类型为多模态数学推理。
This dataset contains approximately 10,000 multimodal mathematical samples, which are employed for training the Vision-R1 model via reinforcement learning. During the GRPO training process, these data are also leveraged to enhance the model's reasoning capabilities. With a total of 10,000 entries, this dataset targets the task of multimodal mathematical reasoning.
提供机构：
Mixed from We-Math, MathVision, Polymath, SceMQA, Geometry3K
