List your company

Toxicity Defense on LLaVA v1.5

22.35Toxicity ScoreNullSteer
19.822436.883753.94571.0063Mar 23, 2026

Evaluation results

MethodLinks
NullSteerepsilon=unconstrained2026.0322.35
ASTRAepsilon=unconstrained2026.0325.1
NullSteerepsilon=32/2552026.0331.82
NullSteerepsilon=16/2552026.0332.55
ASTRAepsilon=32/2552026.0334.76
ASTRAepsilon=16/2552026.0336.02
NullSteerepsilon=64/2552026.0339.32
ASTRAepsilon=64/2552026.0343.13
Jailbreak Templatesepsilon=32/2552026.0350.21
Jailbreak Templatesepsilon=16/2552026.0352.61
Jailbreak Templatesepsilon=unconstrained2026.0354.9
Jailbreak Templatesepsilon=64/2552026.0355.48
Refusal Pairsepsilon=64/2552026.0360.36
Refusal Pairsepsilon=unconstrained2026.0362.46
Refusal Pairsepsilon=16/2552026.0366.72
ECSOepsilon=unconstrained2026.0366.81
Refusal Pairsepsilon=32/2552026.0366.82
ECSOepsilon=64/2552026.0371.32
ECSOepsilon=32/2552026.0373.14
JailGuardepsilon=unconstrained2026.0373.76
ECSOepsilon=16/2552026.0373.77
Benign imageepsilon=16/2552026.0375
Benign imageepsilon=32/2552026.0375
Benign imageepsilon=64/2552026.0375
Benign imageepsilon=unconstrained2026.0375
JailGuardepsilon=64/2552026.0375.76
JailGuardepsilon=16/2552026.0377.6
JailGuardepsilon=32/2552026.0377.77
Self-reminderepsilon=unconstrained2026.0380.93
w/o defenseepsilon=16/2552026.0383.7
Self-reminderepsilon=16/2552026.0383.92
Self-reminderepsilon=32/2552026.0383.97
Self-reminderepsilon=64/2552026.0384.19
w/o defenseepsilon=32/2552026.0384.4
w/o defenseepsilon=unconstrained2026.0385.44
w/o defenseepsilon=64/2552026.0385.54