VisualPuzzles/VisualPuzzles

VisualPuzzles是一个为了评估大型模型的推理能力而特别设计的多模态基准数据集。它包含1168个不同的谜题，分为算法、类比、演绎、归纳和空间五种推理类别，并标注了易、中、难三种难度。与现有的知识密集型基准相比，该数据集的知识密集度较低，而推理复杂度更高。

VisualPuzzles is a multimodal benchmark specifically designed to evaluate the reasoning abilities of large models. It contains 1168 diverse puzzles categorized into five types of reasoning: Algorithmic, Analogical, Deductive, Inductive, and Spatial, with three difficulty levels: Easy, Medium, and Hard. Compared to existing knowledge-intensive benchmarks, this dataset is less knowledge-intensive and more complex in terms of reasoning.