Share your thoughts, 1 month free Claude Pro on usSee more

Unlearning Detection on WMDP

100Accuracy

LLaMA-3.1-8B + Activation-based Detection

Updated 1mo ago

Evaluation Results

Method	Links
LLaMA-3.1-8B + Activation-based Detection 2025.06		100
Yi-34B + Activation-based Detection 2025.06		100
Zephyr-7B 2025.06		100
LLaMA-3.1-8B 2025.06		100
LLaMA-3.1-8B 2025.06		100
Qwen2.5-14B 2025.06		99.86
Yi-34B 2025.06		99.86
Zephyr-7B + Activation-based Detection 2025.06		99.72
Yi-34B 2025.06		99.72
Zephyr-7B 2025.06		99.72
Qwen2.5-14B 2025.06		99.72
Qwen2.5-14B + Activation-based Detection 2025.06		98.59
unlearning trace detection 2025.06		95.07
unlearning trace detection 2025.06		94.37
unlearning trace detection 2025.06		93.24
unlearning trace detection 2025.06		90.56