Harmful prompt detection on WGMix
88.52F1 ScoreMLPM
Evaluation results
| Method | Links | |
|---|---|---|
| MLPMBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 88.52 | |
| Ayub & MajumdarBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 88.09 | |
| MLPMBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 88.04 | |
| WildGuardMethodology=Guard Model2025.02 | 88.04 | |
| MLPMBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 87.63 | |
| Abdelnabi et al.Backbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 86.27 | |
| MLPMBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 86.21 | |
| Abdelnabi et al.Backbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 86.2 | |
| Abdelnabi et al.Backbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 85.73 | |
| Abdelnabi et al.Backbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 85.18 | |
| GraniteGuardian-3-1-8BMethodology=Guard Model2025.02 | 84.57 | |
| Ayub & MajumdarBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 83.31 | |
| Ayub & MajumdarBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 80.91 | |
| Ayub & MajumdarBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 80.64 | |
| LlamaGuard3Methodology=Guard Model2025.02 | 76.76 | |
| Aegis-Guard-DMethodology=Guard Model2025.02 | 72.09 | |
| ShieldGemma-9BMethodology=Guard Model2025.02 | 58.88 |