deca-ai/PackBench
收藏Hugging Face2025-07-08 更新2025-09-13 收录
下载链接:
https://hf-mirror.com/datasets/deca-ai/PackBench
下载链接
链接失效反馈官方服务:
资源简介:
PackBench是一个视觉空间推理任务的测试套件,用于评估语言模型在空间推理、镜像变换和结构化决策制定方面的能力。它由多个选择题或简答题组成,提供清晰的视觉文本说明和示例,使用ASCII艺术形式表示。每个示例包含一个问题提示和一个答案字典。
PackBench is a suite of visual-spatial reasoning tasks designed to evaluate language models abilities in spatial reasoning, mirroring transformations, and structured decision-making. It consists of multiple-choice or short-answer questions with clear visual-textual instructions and examples represented in ASCII art. Each example includes a question prompt and an answer dictionary.
提供机构:
deca-ai



