Code-Vision
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/wanghanbinpanda/CodeVision
下载链接
链接失效反馈官方服务：
资源简介：
该数据集旨在评估多模态大型语言模型（MLLMs）的逻辑理解和代码生成能力，通过根据给定的流程图生成程序来进行基准测试。该数据集不仅覆盖了基本的编程领域，还挑战了MLLMs在算法和数学问题解决方面的能力。数据集分为三个子集：HumanEval-V、Algorithm和MATH，主要任务是基于流程图表示进行代码生成。
This dataset is designed to evaluate the logical comprehension and code generation capabilities of multimodal large language models (MLLMs), with benchmarking carried out by generating programs based on given flowcharts. It covers not only fundamental programming domains, but also tests MLLMs' abilities in solving algorithmic and mathematical problems. The dataset is split into three subsets: HumanEval-V, Algorithm, and MATH, with the primary task being code generation based on flowchart representations.
