List your company

Structural Erasure on WMDP bio

0CADZero-shot
-0.057320.329590.71651.10341May 14, 2026

Evaluation results

MethodLinks
Zero-shotModel=Llama-3.1-8B-Instruct2026.05000
Zero-shotModel=Qwen-3-8B2026.05000
LUNARModel=Qwen-3-8B2026.050.0391.1430.301
LUNARModel=Llama-3.1-8B-Instruct2026.050.0411.1870.312
Surgical GAModel=Llama-3.1-8B-Instruct2026.050.3210.3210.173
Global GAModel=Llama-3.1-8B-Instruct2026.050.5230.5230.311
Surgical GAModel=Qwen-3-8B2026.050.6630.6630.356
GU+SimNPOModel=Qwen-3-8B2026.050.7920.7920.544
NPOModel=Qwen-3-8B2026.050.8470.8470.55
NPOModel=Llama-3.1-8B-Instruct2026.050.8490.8490.509
SimNPOModel=Qwen-3-8B2026.050.9110.9110.609
Global GAModel=Qwen-3-8B2026.051.0211.0210.66
MANSUModel=Qwen-3-8B2026.051.0890.5310.224
MANSUModel=Llama-3.1-8B-Instruct2026.051.1430.4120.138
GU+SimNPOModel=Llama-3.1-8B-Instruct2026.051.2921.2920.824
SimNPOModel=Llama-3.1-8B-Instruct2026.051.4331.4330.87