Prompt Injection Detection on Primary HarmBench + Alpaca (OOD)
75.28Macro-F1DeBERTa-base
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DeBERTa-basebackbone=DeBERTa, variant=base2026.05 | 75.28 | 75.61 | 0.8648 | 0.854 | 0.3368 | 0.5217 | 0.5851 | |
| RoBERTa+IBVS Totalvariant=Total2026.05 | 74.97 | 75.26 | 0.8846 | 0.8674 | 0.4557 | 0.5877 | 0.6328 |