Harmfulness Detection on Aegis 2.0
83.4Macro F1SIREN
Evaluation Results
| Method | Links | |
|---|---|---|
| SIRENBackbone=Qwen3-4B2026.04 | 83.4 | |
| SIRENBackbone=Llama3.1-8B2026.04 | 82.9 | |
| SIRENBackbone=Llama3.2-1B2026.04 | 82.7 | |
| Qwen3GuardBackbone=Qwen3-4B2026.04 | 82.5 | |
| SIRENBackbone=Qwen3-0.6B2026.04 | 82.1 | |
| Qwen3GuardBackbone=Qwen3-0.6B2026.04 | 82 | |
| LlamaGuard3Backbone=Llama3.1-8B2026.04 | 78 | |
| LlamaGuard3Backbone=Llama3.2-1B2026.04 | 72.6 |