FalconCode, Singapore
收藏资源简介:
本研究使用了两个高质量的公开编程数据集:FalconCode和Singapore,这两个数据集被精心挑选用于评估教育程序修复任务。FalconCode数据集包含来自多个学期的学生提交的编程作业,涵盖不同难度级别,特别适合用于评估大型语言模型在程序修复中的应用。Singapore数据集则包含了来自新加坡国立大学的学生编程作业,涉及基础编程概念,适合用于评估模型在无先前数据可用情况下的修复能力。这两个数据集的创建旨在标准化教育程序修复的评估,促进不同模型和方法之间的公平比较,从而推动教育领域中程序修复技术的发展。
This study utilizes two high-quality open programming datasets: FalconCode and Singapore, which were carefully selected for evaluating educational program repair tasks. The FalconCode dataset contains student-submitted programming assignments from multiple semesters, covering varying difficulty levels, making it particularly suitable for evaluating the application of Large Language Models (LLMs) in program repair. The Singapore dataset, on the other hand, comprises student programming assignments from the National University of Singapore (NUS) that cover fundamental programming concepts, and is suitable for evaluating a model's repair capability in scenarios where no prior data is available. The creation of these two datasets aims to standardize the evaluation of educational program repair, facilitate fair comparisons between different models and methods, and thereby advance the development of program repair technologies in the field of education.

- 1Benchmarking Educational Program Repair阿尔托大学 · 2024年



