List your company

Toxicity Defense on MiniGPT-4

2.89Toxicity ScoreNullSteer
0.817614.806328.79542.7837Mar 23, 2026

Evaluation results

MethodLinks
NullSteerepsilon=unconstrained2026.032.89
NullSteerepsilon=64/2552026.033.26
ASTRAepsilon=unconstrained2026.034.48
ASTRAepsilon=64/2552026.034.51
NullSteerepsilon=32/2552026.036.45
ASTRAepsilon=32/2552026.038.84
NullSteerepsilon=16/2552026.039.52
ASTRAepsilon=16/2552026.0311.3
JailGuardepsilon=16/2552026.0316.51
JailGuardepsilon=32/2552026.0318.93
Jailbreak Templatesepsilon=16/2552026.0319.73
JailGuardepsilon=64/2552026.0320.93
JailGuardepsilon=unconstrained2026.0321.23
Jailbreak Templatesepsilon=unconstrained2026.0322.78
Jailbreak Templatesepsilon=32/2552026.0325.03
Refusal Pairsepsilon=16/2552026.0325.76
Jailbreak Templatesepsilon=64/2552026.0330.1
Refusal Pairsepsilon=32/2552026.0330.28
Benign imageepsilon=16/2552026.0330.65
Benign imageepsilon=32/2552026.0330.65
Benign imageepsilon=64/2552026.0330.65
Benign imageepsilon=unconstrained2026.0330.65
Refusal Pairsepsilon=64/2552026.0331.99
ECSOepsilon=32/2552026.0332.42
ECSOepsilon=16/2552026.0334.59
Refusal Pairsepsilon=unconstrained2026.0335.71
ECSOepsilon=64/2552026.0338.54
Self-reminderepsilon=16/2552026.0338.97
w/o defenseepsilon=16/2552026.0339.73
ECSOepsilon=unconstrained2026.0342.86
Self-reminderepsilon=64/2552026.0345.15
w/o defenseepsilon=32/2552026.0348.52
Self-reminderepsilon=32/2552026.0348.71
Self-reminderepsilon=unconstrained2026.0350.12
w/o defenseepsilon=unconstrained2026.0352.12
w/o defenseepsilon=64/2552026.0354.7