Knowledge Unlearning on WMDP cyber
47.21AccuracyOriginal model
Evaluation results
| Method | Links | |
|---|---|---|
| Original modelUnlearning Dataset=N/A2026.01 | 47.21 | |
| RMUUnlearning Dataset=Filter2026.01 | 46.9 | |
| ELMUnlearning Dataset=Keyword2026.01 | 46.3 | |
| ELMUnlearning Dataset=Filter2026.01 | 46.25 | |
| BaseBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 45.3 | |
| NPO_KLBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 45.3 | |
| ELMUnlearning Dataset=Textbook2026.01 | 45.29 | |
| BaseBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 44.3 | |
| NPOBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 43.6 | |
| NPO_GDBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 43.1 | |
| NPO_KLBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 41.3 | |
| RMUUnlearning Dataset=Keyword2026.01 | 38.2 | |
| RMUBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 37 | |
| RefBackbone=StableLM Zephyr 3B2026.04 | 35.78 | |
| NPOBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 34.4 | |
| ELMUnlearning Dataset=BiForget2026.01 | 33.87 | |
| NPO_GDBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 33.1 | |
| LoRABackbone=Llama3-8B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 32.1 | |
| LoRABackbone=Zephyr-7B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 32.1 | |
| OBLIVATEUnlearning Dataset=Filter2026.01 | 31.58 | |
| SCNBackbone=StableLM Zephyr 3B2026.04 | 31.25 | |
| SatImpBackbone=StableLM Zephyr 3B2026.04 | 30.95 | |
| OBLIVATEUnlearning Dataset=Keyword2026.01 | 30.55 | |
| NPOBackbone=StableLM Zephyr 3B2026.04 | 30.05 | |
| TNPOBackbone=StableLM Zephyr 3B2026.04 | 29.94 | |
| WGABackbone=StableLM Zephyr 3B2026.04 | 29.89 | |
| GABackbone=StableLM Zephyr 3B2026.04 | 29.79 | |
| ETWBackbone=StableLM Zephyr 3B2026.04 | 29.79 | |
| AsymLoRABackbone=Llama3-8B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 28.8 | |
| SCEBackbone=StableLM Zephyr 3B2026.04 | 28.64 | |
| RMUUnlearning Dataset=BiForget2026.01 | 28.58 | |
| OBLIVATEUnlearning Dataset=Textbook2026.01 | 27.98 | |
| RMUBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 27.3 | |
| RMUUnlearning Dataset=Textbook2026.01 | 27.22 | |
| ELMBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 27.2 | |
| ELMUnlearning Dataset=Official2026.01 | 27.13 | |
| GABackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 26.8 | |
| ELMBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 26.6 | |
| RLBackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 26.6 | |
| OBLIVATEUnlearning Dataset=BiForget2026.01 | 26.52 | |
| RMUUnlearning Dataset=Official2026.01 | 26.32 | |
| AsymLoRABackbone=Zephyr-7B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 26.3 | |
| OBLIVATEUnlearning Dataset=Official2026.01 | 25.72 | |
| ALTERBackbone=Llama3-8B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 25.6 | |
| RLBackbone=Zephyr-7B, Unlearning Framework=FT & Model Editing2026.03 | 24.7 | |
| GABackbone=Llama3-8B, Unlearning Framework=FT & Model Editing2026.03 | 24 | |
| ALTERBackbone=Zephyr-7B, Unlearning Framework=LoRA Variations(r=8)2026.03 | 24 |