PKU-Alignment/s1-m_beta

S1-M Dataset (Beta)是一个用于训练具有“先思考，后回答”范式的S1-M模型的开源TI2T推理数据集。该数据集包含图片和对应的提示、回答，以及来源类型信息。图片和提示来源于两个数据集：align-anything和multimodal-open-r1-8k-verified，旨在平衡模型的一般能力和数学能力。数据集的注释使用了Claude 3.7 Sonnet 20250219模型，按照系统提示进行思考并给出答案。

S1-M Dataset (Beta) is an open-source TI2T reasoning dataset for training the S1-M Model with a think first, then respond paradigm. The dataset includes images along with corresponding prompts, responses, and source type information. The images and prompts are sourced from two datasets: align-anything and multimodal-open-r1-8k-verified, aiming to balance the models general capabilities with mathematical abilities. The dataset annotation uses the Claude 3.7 Sonnet 20250219 model, which is guided by system prompts to think and then provide answers.