Safety Auditing on Llama-3.2-11B-Vision-Instruct (test)
41.85Detoxify ScoreRedDiff w/ alignment (L)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RedDiff w/ alignment (L)Modality=Multimodal, Alignment level=L2025.03 | 41.85 | 45.24 | |
| RedDiff w/ alignment (M)Modality=Multimodal, Alignment level=M2025.03 | 41.81 | 44.37 | |
| RedDiff w/ alignment (H)Modality=Multimodal, Alignment level=H2025.03 | 41.15 | 43.78 | |
| RedDiff w/o alignmentModality=Multimodal, Alignment=None2025.03 | 41.01 | 43.9 | |
| Text-only BaselineModality=Text-only2025.03 | 15.02 | 18.68 |