Multimodal Safety Evaluation on MM-SafeBench
1.04Forbidden Statements ASRGPT-4o
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GPT-4oEvaluator=Llama-Guard-3-8B, System Prompt=None2024.11 | 1.04 | 13.23 | 95.21 | 95.48 | 95.91 | 96.4 | |
| Claude-3.5-SonnetEvaluator=Llama-Guard-3-8B, System Prompt=None2024.11 | 2.44 | 2.78 | 40.02 | 48.62 | 37.12 | 9.28 | |
| GPT-4o-MiniEvaluator=Llama-Guard-3-8B, System Prompt=None2024.11 | 16.13 | 12.06 | 94.78 | 94.78 | 93.16 | 93.16 | |
| Qwen-VL-MaxEvaluator=Llama-Guard-3-8B, System Prompt=None2024.11 | 27.84 | 48.49 | 91.76 | 92.34 | 91.42 | 92.23 |