List your company

Unlearning Detection on WMDP

100AccuracyLLaMA-3.1-8B + Activation-based Detection
90.182492.731295.2897.8288Jun 16, 2025

Evaluation results

MethodLinks
LLaMA-3.1-8B + Activation-based DetectionModel=LLaMA-3.1-8B, Unlearning method=RMU, Feature type=RMS-normalized final-layer activations, Classifier=two-layer MLP, Unlearning target=WMDP2025.06100
Yi-34B + Activation-based DetectionModel=Yi-34B, Unlearning method=RMU, Feature type=RMS-normalized final-layer activations, Classifier=two-layer MLP, Unlearning target=WMDP2025.06100
Zephyr-7BUnlearning Method=NPO, Input Features=RMS-normalized final-layer activations, Unlearning Target Dataset=WMDP2025.06100
LLaMA-3.1-8BUnlearning Method=NPO, Input Features=RMS-normalized final-layer activations, Unlearning Target Dataset=WMDP2025.06100
LLaMA-3.1-8BUnlearning method=NPO, Detection source=Text-based2025.06100
Qwen2.5-14BUnlearning Method=NPO, Input Features=RMS-normalized final-layer activations, Unlearning Target Dataset=WMDP2025.0699.86
Yi-34BUnlearning method=NPO, Detection source=Text-based2025.0699.86
Zephyr-7B + Activation-based DetectionModel=Zephyr-7B, Unlearning method=RMU, Feature type=RMS-normalized final-layer activations, Classifier=two-layer MLP, Unlearning target=WMDP2025.0699.72
Yi-34BUnlearning Method=NPO, Input Features=RMS-normalized final-layer activations, Unlearning Target Dataset=WMDP2025.0699.72
Zephyr-7BUnlearning method=NPO, Detection source=Text-based2025.0699.72
Qwen2.5-14BUnlearning method=NPO, Detection source=Text-based2025.0699.72
Qwen2.5-14B + Activation-based DetectionModel=Qwen2.5-14B, Unlearning method=RMU, Feature type=RMS-normalized final-layer activations, Classifier=two-layer MLP, Unlearning target=WMDP2025.0698.59
unlearning trace detectionModel=Qwen2.5-14B2025.0695.07
unlearning trace detectionModel=Yi-34B2025.0694.37
unlearning trace detectionModel=LLaMA-3.1-8B2025.0693.24
unlearning trace detectionModel=Zephyr-7B2025.0690.56