JetBrains/git_good_bench-train

GitGoodBench Lite是一个包含17469个样本的子集，用于收集AI代理解决Git任务的轨迹（见支持的场景）以供模型训练使用。我们支持Python、Java和Kotlin编程语言以及合并冲突解决和文件提交链的样本类型。数据集中的所有数据都是从符合特定条件的816个唯一的开源GitHub仓库中收集的。数据集包含两种类型的样本：合并和文件提交链。数据集的结构详细描述了每个字段的类型和描述，以及是否为元数据。此外，README还提供了关于仓库、编程语言和合并冲突解决样本的多样性统计信息。

GitGoodBench Lite is a subset of 17469 samples for collecting trajectories of AI agents resolving git tasks (see Supported Scenarios) for model training purposes. It supports programming languages Python, Java, and Kotlin, and sample types including merge conflict resolution and file-commit chain. All data are collected from 816 unique, open-source GitHub repositories under specific conditions. The dataset contains two types of samples: merge and file_commit_chain. The dataset structure details each fields type, description, and whether it is metadata. Additionally, the README provides statistics on the diversity of repositories, programming languages, and merge conflict resolution samples.