Safety Evaluation on JailbreakLLMs Orig.
0Unsafe RateVLGuard PH
Evaluation Results
| Method | Links | |
|---|---|---|
| VLGuard PHBackbone=LLaVA-v1.5-13B2024.10 | 0 | |
| VLGuard MixedBackbone=LLaVA-v1.5-13B2024.10 | 0.38 | |
| VLGuard MixedBackbone=LLaVA-v1.5-7B2024.10 | 0.76 | |
| CMRM_sampleBackbone=ShareGPT4V2024.10 | 1.14 | |
| LLaVA-v1.5-13BInput Protocol=Caption2024.10 | 1.82 | |
| VLGuard PHBackbone=LLaVA-v1.5-7B2024.10 | 2.27 | |
| LLaVA-v1.5-13BInput Protocol=Query2024.10 | 3.03 | |
| CMRM_sampleBackbone=LLaVA-v1.5-13B2024.10 | 3.03 | |
| CMRM_datasetBackbone=ShareGPT4V2024.10 | 3.79 | |
| CMRM_sampleBackbone=LLaVA-v1.5-7B2024.10 | 4.55 | |
| LLaVA-v1.5-7BInput Protocol=Caption2024.10 | 4.85 | |
| CMRM_datasetBackbone=LLaVA-v1.5-13B2024.10 | 4.92 | |
| CMRM_datasetBackbone=LLaVA-v1.5-7B2024.10 | 8.33 | |
| ShareGPT4VInput Protocol=Caption2024.10 | 8.33 | |
| ShareGPT4VInput Protocol=Query2024.10 | 10.23 | |
| LLaVA-v1.5-13BInput Protocol=Visual (Orig.)2024.10 | 12.42 | |
| LLaVA-v1.5-7BInput Protocol=Query2024.10 | 12.73 | |
| ShareGPT4VInput Protocol=Visual (Orig.)2024.10 | 19.32 | |
| LLaVA-v1.5-7BInput Protocol=Visual (Orig.)2024.10 | 21.52 |