Prompt Injection Detection on In-distribution (ID) (test)
95.41Macro F1 ScoreRoBERTa+IBVS Total
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RoBERTa+IBVS Totalvariant=Total2026.05 | 95.41 | 96.42 | 99.23 | 98.25 | 56.33 | 95.99 | 98.46 | |
| RoBERTa+IBVS TotalRegime=ID2026.05 | 95.41 | 96.42 | 99.23 | 98.25 | 56.33 | 95.99 | 98.46 | |
| DeBERTa-basebackbone=DeBERTa, variant=base2026.05 | 93.46 | 94.86 | 99.44 | 98.62 | 74.42 | 94.46 | 97.23 | |
| DeBERTa-baseRegime=ID2026.05 | 93.46 | 94.86 | 99.44 | 98.62 | 74.42 | 94.46 | 97.23 | |
| LLM Guard PromptInjectionRegime=ID, Classifier=ProtectAI deberta-v3-base-prompt-injection-v22026.05 | 32.21 | 34.68 | 90.19 | 80.19 | 11.09 | 27.39 | 46.5 |