SWE-Swiss/SWESwiss-SFT-Repair-4K

SWE-Swiss 数据集用于训练 SWE-Swiss 模型进行修复任务。该数据集的 prompts 基于来自 SWE-Gym 和 SWE-smith 的 issues。每个 prompt 中的代码内容包含两个部分：oracle 文件，即需要打补丁的地面真实文件；以及 distractor 文件，即由 LLM 预测的看似合理但错误的文件。responses 由 DeepSeek-R1-0528 生成，并且我们过滤掉了生成的补丁无法通过仓库单元测试的数据。The SWE-Swiss dataset is used for training SWE-Swiss models on the repair task. The prompts are based on issues from SWE-Gym and SWE-smith. Each prompts code content consists of two parts: oracle files, which are the ground-truth files that require a patch, and distractor files, which are plausible but incorrect files predicted by an LLM. The responses are generated by DeepSeek-R1-0528, and we filter out any data where the generated patch fails to pass the repositorys unit tests.