List your company

Knowledge Retention on WMDP bio (retain)

80.3RtZero-shot
18.62834.63950.6566.661May 14, 2026

Evaluation results

MethodLinks
Zero-shotBackbone=Qwen-3-8B2026.0580.3
Zero-shotBackbone=Llama-3.1-8B-Instruct2026.0576.3
MANSUBackbone=Qwen-3-8B2026.0567.1
LUNARBackbone=Qwen-3-8B2026.0565.5
LUNARBackbone=Llama-3.1-8B-Instruct2026.0561.9
Surgical GABackbone=Llama-3.1-8B-Instruct2026.0556
MANSUBackbone=Llama-3.1-8B-Instruct2026.0552.3
NPOBackbone=Llama-3.1-8B-Instruct2026.0550.3
Surgical GABackbone=Qwen-3-8B2026.0530.3
NPOBackbone=Qwen-3-8B2026.0530.3
GU+SimNPOBackbone=Qwen-3-8B2026.0527.7
Global GABackbone=Llama-3.1-8B-Instruct2026.0526
SimNPOBackbone=Qwen-3-8B2026.0525.7
GU+SimNPOBackbone=Llama-3.1-8B-Instruct2026.0524.7
Global GABackbone=Qwen-3-8B2026.0524.7
SimNPOBackbone=Llama-3.1-8B-Instruct2026.0521