Structural Erasure on WMDP bio
0CADZero-shot
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Zero-shotModel=Llama-3.1-8B-Instruct2026.05 | 0 | 0 | 0 | |
| Zero-shotModel=Qwen-3-8B2026.05 | 0 | 0 | 0 | |
| LUNARModel=Qwen-3-8B2026.05 | 0.039 | 1.143 | 0.301 | |
| LUNARModel=Llama-3.1-8B-Instruct2026.05 | 0.041 | 1.187 | 0.312 | |
| Surgical GAModel=Llama-3.1-8B-Instruct2026.05 | 0.321 | 0.321 | 0.173 | |
| Global GAModel=Llama-3.1-8B-Instruct2026.05 | 0.523 | 0.523 | 0.311 | |
| Surgical GAModel=Qwen-3-8B2026.05 | 0.663 | 0.663 | 0.356 | |
| GU+SimNPOModel=Qwen-3-8B2026.05 | 0.792 | 0.792 | 0.544 | |
| NPOModel=Qwen-3-8B2026.05 | 0.847 | 0.847 | 0.55 | |
| NPOModel=Llama-3.1-8B-Instruct2026.05 | 0.849 | 0.849 | 0.509 | |
| SimNPOModel=Qwen-3-8B2026.05 | 0.911 | 0.911 | 0.609 | |
| Global GAModel=Qwen-3-8B2026.05 | 1.021 | 1.021 | 0.66 | |
| MANSUModel=Qwen-3-8B2026.05 | 1.089 | 0.531 | 0.224 | |
| MANSUModel=Llama-3.1-8B-Instruct2026.05 | 1.143 | 0.412 | 0.138 | |
| GU+SimNPOModel=Llama-3.1-8B-Instruct2026.05 | 1.292 | 1.292 | 0.824 | |
| SimNPOModel=Llama-3.1-8B-Instruct2026.05 | 1.433 | 1.433 | 0.87 |