jordan-taylor-aisi/wmdp_bio_cot_prompted_sandbagging_llama_31_8b_instruct
收藏Hugging Face2025-04-09 更新2025-04-12 收录
下载链接:
https://hf-mirror.com/datasets/jordan-taylor-aisi/wmdp_bio_cot_prompted_sandbagging_llama_31_8b_instruct
下载链接
链接失效反馈官方服务:
资源简介:
这是一个用于评估和检测沙袋策略的数据集,包含了deception_sprint/wmdp_bio任务的数据。数据集使用了vllm/meta-llama/Llama-3.1-8B-Instruct模型,并提供了良性提示和恶意提示来指导模型行为。数据集创建于2025年4月9日,启用了沙袋策略检测但没有过滤。数据集没有分割,但有指定测试和验证集的大小。
This is a dataset for evaluating and detecting sandbagging strategies, containing data for the deception_sprint/wmdp_bio task. The dataset uses the vllm/meta-llama/Llama-3.1-8B-Instruct model and provides both benign and malicious prompts to guide model behavior. The dataset was created on April 9, 2025, with sandbagging detection enabled but no filtering applied. The dataset is not split, but test and validation set sizes are specified.
提供机构:
jordan-taylor-aisi



