Safety Evaluation on JailbreakLLMs Noise
0.38Unsafe RateVLGuard Mixed
Evaluation Results
| Method | Links | |
|---|---|---|
| VLGuard MixedBackbone=LLaVA-v1.5-7B2024.10 | 0.38 | |
| VLGuard MixedBackbone=LLaVA-v1.5-13B2024.10 | 0.38 | |
| VLGuard PHBackbone=LLaVA-v1.5-13B2024.10 | 0.38 | |
| CMRM_sampleBackbone=LLaVA-v1.5-13B2024.10 | 2.27 | |
| VLGuard PHBackbone=LLaVA-v1.5-7B2024.10 | 2.65 | |
| CMRM_sampleBackbone=LLaVA-v1.5-7B2024.10 | 4.92 | |
| CMRM_sampleBackbone=ShareGPT4V2024.10 | 6.06 | |
| CMRM_datasetBackbone=LLaVA-v1.5-7B2024.10 | 7.2 | |
| CMRM_datasetBackbone=ShareGPT4V2024.10 | 8.33 | |
| CMRM_datasetBackbone=LLaVA-v1.5-13B2024.10 | 8.71 | |
| LLaVA-v1.5-13BInput Protocol=Noisy Image2024.10 | 14.77 | |
| ShareGPT4VInput Protocol=Noisy Image2024.10 | 24.24 | |
| LLaVA-v1.5-7BInput Protocol=Noisy Image2024.10 | 25.76 |