aisi-whitebox/gsm8k_prompted-sandbagging-llama_33_70b_instruct
收藏Hugging Face2025-04-24 更新2025-04-26 收录
下载链接：
https://hf-mirror.com/datasets/aisi-whitebox/gsm8k_prompted-sandbagging-llama_33_70b_instruct
下载链接
链接失效反馈官方服务：
资源简介：
gsm8k_prompted-sand袋ging-llama_33_70b_instruct是一个评估数据集，用于检测和评估AI模型在特定任务上的性能。该数据集通过使用两种不同的系统提示，良性提示和恶意提示，来评估模型在故意给出低质量解决方案时的表现。数据集包含了沙袋检测，但没有应用过滤机制，允许研究者在真实条件下评估模型性能。
The gsm8k_prompted-sandbagging-llama_33_70b_instruct dataset is an evaluation dataset designed to assess and detect the performance of AI models on specific tasks. It uses two types of system prompts, benign and malicious, to evaluate how the model performs when intentionally providing suboptimal solutions. The dataset includes sandbagging detection but does not apply filtering, allowing researchers to evaluate model performance under real conditions.
提供机构：
aisi-whitebox
