Win rate evaluation on SLIMORCA (test)
88.12Win RateTOSS
Evaluation Results
| Method | Links | |
|---|---|---|
| TOSSBackbone=Llama-3-8B-Instruct2026.03 | 88.12 | |
| TOSSComparison model=Standard SFT, Backbone=Llama-3-8B-Instruct2026.03 | 68.37 | |
| TOSSParadigm=Training-based defense, Base Model=Llama-3-8B-Instruct2026.03 | 68.37 | |
| TOSSComparison model=SEAL, Backbone=Llama-3-8B-Instruct2026.03 | 63.96 | |
| Circuit BreakersParadigm=Training-based defense, Base Model=Llama-3-8B-Instruct2026.03 | 50.95 | |
| SafetyLockParadigm=Training-free defense, Base Model=Llama-3-8B-Instruct2026.03 | 50.72 | |
| Standard SFTBackbone=Llama-3-8B-Instruct2026.03 | 50 | |
| Standard SFTParadigm=/, Base Model=Llama-3-8B-Instruct2026.03 | 50 |