jprivera44/exp2_transcript_monitor_results
收藏Hugging Face2025-12-18 更新2025-12-20 收录
下载链接：
https://hf-mirror.com/datasets/jprivera44/exp2_transcript_monitor_results
下载链接
链接失效反馈官方服务：
资源简介：
该数据集用于训练和测试模型以检测代码中的后门，包括带有细微信号的类型。实验通过微调模型来识别特定的共谋信号（如`# type: ignore`），并展示了模型在检测未训练信号类型时的泛化能力。数据集包含平衡的训练和测试样本，涵盖干净代码、无信号后门以及带有不同信号标记的后门代码。
This dataset is used for training and testing models to detect backdoors in code, including those with subtle signals. The experiment involves fine-tuning a model to recognize specific collusion signals (e.g., `# type: ignore`) and demonstrates the models generalization ability in detecting untrained signal types. The dataset includes balanced training and test samples, covering clean code, backdoors without signals, and backdoor code with different signal markers.
提供机构：
jprivera44
