SWE-Swiss/SWESwiss-SFT-Localization-5K

SWE-Swiss数据集用于训练SWE-Swiss模型进行本地化任务。该数据集的提示是从SWE-Gym-Raw数据集和一个SWE-bench训练集的子集中构建的。为了防止数据泄露，过滤掉了在SWE-bench测试集中出现的任何仓库。响应是由DeepSeek-R1-0528模型生成的。只有当模型的预测满足两个条件时，一个实例才会被包含在最终数据集中：预测的文件数量不超过五个，且召回率为1.0（即正确识别所有真实文件）。

The SWE-Swiss dataset is used for training SWE-Swiss models on the localization task. The prompts are constructed from a subset of issues in the SWE-Gym-Raw dataset and the SWE-bench training set. Any repositories that also appear in the SWE-bench test set have been filtered out to prevent data leakage. The responses are generated by the DeepSeek-R1-0528 model. An instance is included in the final dataset only if the models prediction meets two conditions: the number of predicted files is five or fewer, and the recall rate is 1.0 (meaning all ground-truth files are correctly identified).