List your company

Mathematical Reasoning on GSM8K (UR, AM, SM, MS)

1.02Utility Ratio (UR)Direct
0.01120.27310.5350.7969May 16, 2025

Evaluation results

MethodLinks
DirectVictim Model=L-70B2025.051.02110.96
FakeVictim Model=P-3.8B2025.051.01110.95
No DefenseVictim Model=L-70B2025.051110.96
GuardVictim Model=L-70B2025.051110.96
No DefenseVictim Model=L-8B2025.051110.96
DirectVictim Model=L-8B2025.051110.96
GuardVictim Model=L-8B2025.051110.96
ProxyPrompt†Victim Model=L-8B2025.05100.130.23
No DefenseVictim Model=P-3.8B2025.0510.7510.95
DirectVictim Model=P-3.8B2025.0510.3810.9
GuardVictim Model=P-3.8B2025.0510.7510.95
ProxyPrompt†Victim Model=P-3.8B2025.05100.250.36
ProxyPromptVictim Model=L-70B2025.050.99000.17
ProxyPromptVictim Model=L-8B2025.050.99000.18
ProxyPromptVictim Model=P-3.8B2025.050.99000.18
ProxyPrompt†Victim Model=L-70B2025.050.98000.2
FakeVictim Model=L-8B2025.050.98110.95
FakeVictim Model=L-70B2025.050.97110.96
FilterVictim Model=P-3.8B2025.050.9500.130.36
FilterVictim Model=L-70B2025.050.38110.91
FilterVictim Model=L-8B2025.050.050.880.880.72