MMMU-Pro
收藏资源简介:
MMMU-Pro是由MMMU团队创建的多学科多模态理解与推理基准数据集,包含3460个精心策划的多模态问题,涵盖六个核心学科。数据集通过过滤可由纯文本模型回答的问题、增加候选选项和引入仅视觉输入设置,严格评估模型的多模态理解和推理能力。创建过程中,数据集通过人工验证和多样化的视觉输入设置,确保问题的高质量和挑战性。MMMU-Pro主要应用于评估和提升多模态AI模型的理解和推理能力,旨在解决当前模型在多模态任务中的局限性。
MMMU-Pro is a multidisciplinary multimodal understanding and reasoning benchmark dataset developed by the MMMU team, which comprises 3,460 carefully curated multimodal questions spanning six core disciplines. To strictly evaluate the multimodal understanding and reasoning capabilities of AI models, this dataset filters out questions that can be answered solely by plain text models, supplements additional candidate options, and introduces visual-only input settings. During the construction process, manual verification and diverse visual input configurations are employed to guarantee the high quality and challenging nature of all questions. MMMU-Pro is primarily utilized to evaluate and enhance the understanding and reasoning abilities of multimodal AI models, with the objective of addressing the current limitations faced by such models in multimodal tasks.

- 1MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding BenchmarkMMMU团队 · 2024年



