CommitBench
收藏资源简介:
CommitBench是由哈索·普拉特纳软件研究所创建的一个大规模数据集,用于训练和评估AI模型生成提交信息。该数据集包含1,664,590条高质量提交信息,来源于38,578个组织和71,676个独特项目,涵盖Java、Python、Go、JavaScript、PHP和Ruby等20种编程语言。数据集通过严格的过滤和增强技术确保数据质量,考虑了隐私、可重现性和许可证等因素。CommitBench旨在解决现有数据集在提交信息生成任务中的不足,提供一个全面的、精心策划的、关注隐私的数据集,以作为未来研究的标准基准。
CommitBench is a large-scale dataset created by the Hasso Plattner Institute for Software Systems, designed for training and evaluating AI models to generate commit messages. This dataset contains 1,664,590 high-quality commit messages sourced from 38,578 organizations and 71,676 unique projects, covering 20 programming languages including Java, Python, Go, JavaScript, PHP, Ruby and others. The dataset ensures data quality through rigorous filtering and enhancement techniques, with considerations for factors such as privacy, reproducibility and licenses. CommitBench aims to address the limitations of existing datasets in the commit message generation task, providing a comprehensive, carefully curated, privacy-focused dataset to serve as a standard benchmark for future research.

- 1CommitBench: A Benchmark for Commit Message Generation哈索·普拉特纳软件研究所 · 2024年



