Harmful prompt detection on WJB
97.55F1 ScoreMLPM
Evaluation results
| Method | Links | |
|---|---|---|
| MLPMBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 97.55 | |
| WildGuardMethodology=Guard Model2025.02 | 97.1 | |
| Ayub & MajumdarBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 96.84 | |
| GraniteGuardian-3-1-8BMethodology=Guard Model2025.02 | 96.75 | |
| MLPMBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 94.69 | |
| MLPMBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 93.65 | |
| Abdelnabi et al.Backbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 93.48 | |
| Abdelnabi et al.Backbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 93.27 | |
| MLPMBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 92 | |
| Abdelnabi et al.Backbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 91.13 | |
| Abdelnabi et al.Backbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 90.44 | |
| Ayub & MajumdarBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 87.53 | |
| Ayub & MajumdarBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 82.85 | |
| Ayub & MajumdarBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 81.31 | |
| Aegis-Guard-DMethodology=Guard Model2025.02 | 75.44 | |
| LlamaGuard3Methodology=Guard Model2025.02 | 67.83 | |
| ShieldGemma-9BMethodology=Guard Model2025.02 | 59.94 |