modelscope/MMMU-Reasoning-Distill-Validation
收藏Hugging Face2025-03-19 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/modelscope/MMMU-Reasoning-Distill-Validation
下载链接
链接失效反馈官方服务：
资源简介：
MMMU-Reasoning-Distill-Validation是一个多模态推理数据集，包含839个图像描述和自然语言推理数据样本，基于MMMU验证集构建。数据集使用Qwen2.5-VL-72B-Instruct模型进行图像理解和生成详细图像描述，然后使用DeepSeek-R1模型生成推理对话。数据集适用于探索VL模型的推理能力，并支持单图像推理和多图像推理。
MMMU-Reasoning-Distill-Validation is a multi-modal reasoning dataset containing 839 image descriptions and natural language inference data samples, built upon the validation set of MMMU. The dataset uses the Qwen2.5-VL-72B-Instruct model for image understanding and generating detailed image descriptions, followed by the DeepSeek-R1 model to generate reasoning conversations. It is suitable for exploring the reasoning capabilities of VL models and supports both single-image reasoning and multi-image reasoning.
提供机构：
modelscope
