SafetyBench
Benchmarks
TaskDatasetSOTA resultTrendResults
SafetyBench English (test)
60.13Accuracy
35
SafetyBench
89Accuracy
28
SafetyBench
69.4Safety
26
SafetyBench en
81.2Avg Score
25
SafetyBench zh
83.2Avg Score
21
SafetyBench MCV
79.79ASR (1-Clip)
16
SafetyBench (test)
81.321Accuracy
9
SafetyBench (test)
89.3Macro F1-Score
8
SafetyBench LLaVA-2 Integrated from AdvBench (test)
83.73Illegal Activity Success Rate
4
SafetyBench MiniGPT-4 Integrated from AdvBench (test)
0.702IA (Illegal Activity)
4