Code2Logic/GameQA-140K

GameQA是一个包含约140,000个问答对的大规模、多样化的模态推理数据集，涵盖30种独特的游戏和158种不同的任务，覆盖了各种认知技能。数据集格式为视觉问答(VQA)，包括游戏状态图像、针对性问题、逐步推理和最终答案。问题类型包括多项选择和填空题。该数据集对于当前最先进的VLMs来说具有挑战性，难度分为图像复杂度级别和任务复杂度级别。

GameQA is a large-scale, diverse modal reasoning dataset containing approximately 140,000 question-answer pairs, covering 30 unique games and 158 distinct tasks that span various cognitive skills. The dataset is formatted as Visual Question Answering (VQA), including game state images, targeted questions, step-by-step reasoning, and final answers. Question types include multiple-choice and fill-in-the-blank. The dataset poses a challenge for state-of-the-art VLMs, with difficulty levels categorized by image complexity and task complexity.