ClassEval-T
收藏资源简介:
ClassEval-T是一个面向类级别的代码翻译基准数据集,由山东大学构建。该数据集包含Python、Java和C++三种编程语言的类级别代码片段,旨在评估大型语言模型在实际开发中的代码翻译能力。数据集通过360人小时的逐行手动翻译完成,包含丰富的依赖信息和实际编程问题,如数据库操作和游戏设计。创建过程中,研究人员设计了三种翻译策略,并评估了六种不同类型的大型语言模型。ClassEval-T的应用领域主要集中在代码翻译和大型语言模型的性能评估,旨在解决现有基准数据集在实际开发中代码翻译能力的不足。
ClassEval-T is a class-level code translation benchmark dataset constructed by Shandong University. It encompasses class-level code snippets across three programming languages: Python, Java, and C++, and aims to evaluate the code translation capabilities of large language models (LLMs) in real-world software development. The dataset was completed through 360 person-hours of line-by-line manual translation, and includes rich dependency information as well as practical programming scenarios such as database operations and game design. During its development, researchers designed three translation strategies and evaluated six distinct types of large language models. ClassEval-T is primarily applied in the fields of code translation and LLM performance evaluation, and is intended to address the shortcomings of existing benchmark datasets in evaluating code translation capabilities during practical software development.




