RepoTransBench
收藏资源简介:
RepoTransBench是一个大规模的真实世界代码翻译基准数据集,由中山大学等机构创建。该数据集包含100个真实代码仓库样本,旨在评估从一种编程语言到另一种编程语言的完整代码仓库翻译能力。数据集的创建过程包括从The Stack和The Stack v2中收集真实仓库,筛选符合条件的仓库,并通过LLM和人工协作构建自动测试套件。RepoTransBench的应用领域主要集中在代码翻译工具的性能评估,旨在解决现有细粒度代码翻译基准无法准确反映真实开发需求的问题。
RepoTransBench is a large-scale real-world code translation benchmark dataset created by Sun Yat-sen University and other institutions. It contains 100 real code repository samples, aiming to evaluate the full code repository translation capability from one programming language to another. The dataset construction process includes collecting real repositories from The Stack and The Stack v2, screening eligible repositories, and building automated test suites via collaborative efforts between LLMs and human annotators. Its primary application scenario is the performance evaluation of code translation tools, which aims to address the problem that existing fine-grained code translation benchmarks cannot accurately reflect real-world development needs.

- 1RepoTransBench: A Real-World Benchmark for Repository-Level Code Translation中山大学 · 2024年



