Harmful prompt detection on SimpST
100F1 ScoreAyub & Majumdar
Evaluation results
| Method | Links | |
|---|---|---|
| Ayub & MajumdarBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 100 | |
| MLPMBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 100 | |
| PolyGuard-QwenSize=7B2026.07 | 100 | |
| HaloGuard 1.0Size=0.8B2026.07 | 100 | |
| HaloGuard 1.0Size=4B2026.07 | 100 | |
| MLPMBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 99.5 | |
| MLPMBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 99.5 | |
| Abdelnabi et al.Backbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 99.5 | |
| LlamaGuard3Methodology=Guard Model2025.02 | 99.5 | |
| GraniteGuardian-3-1-8BMethodology=Guard Model2025.02 | 99.5 | |
| WildGuardMethodology=Guard Model2025.02 | 99.5 | |
| WildGuardSize=7B2026.07 | 99.5 | |
| Qwen3Guard-GenSize=4B2026.07 | 99.5 | |
| Qwen3Guard-GenSize=8B2026.07 | 99.5 | |
| Qwen3Guard-GenSize=0.6B2026.07 | 99 | |
| Abdelnabi et al.Backbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 98.99 | |
| Ayub & MajumdarBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 98.99 | |
| NemoGuardSize=8B2026.07 | 98.5 | |
| Ayub & MajumdarBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 98.48 | |
| Abdelnabi et al.Backbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 98.48 | |
| LlamaGuard4Size=12B2026.07 | 98 | |
| Abdelnabi et al.Backbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 97.96 | |
| Aegis-Guard-DMethodology=Guard Model2025.02 | 97.96 | |
| MLPMBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 96.91 | |
| Ayub & MajumdarBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 95.29 | |
| ShieldGemma-9BMethodology=Guard Model2025.02 | 91.3 | |
| ShieldGemmaSize=27B2026.07 | 84.4 |