Safety Evaluation on Safety Tweet Eval, Hatecheck, Ethos (test)
83.3AccuracyContext Distill.
Evaluation results
| Method | Links | |
|---|---|---|
| Context Distill.Model=Llama-3.2-3B-Ins2026.02 | 83.3 | |
| OPCDModel=Llama-3.2-3B-Ins2026.02 | 83.1 | |
| OPCDModel=Llama-3.1-8B-Ins2026.02 | 79.6 | |
| OPCDModel=Qwen2.5-7B-Ins2026.02 | 78.1 | |
| Context Distill.Model=Llama-3.1-8B-Ins2026.02 | 77.2 | |
| Context Distill.Model=Qwen2.5-7B-Ins2026.02 | 77 | |
| In-ContextModel=Llama-3.1-8B-Ins2026.02 | 75.3 | |
| In-ContextModel=Qwen2.5-7B-Ins2026.02 | 72.7 | |
| Base ModelModel=Llama-3.1-8B-Ins2026.02 | 70.7 | |
| In-ContextModel=Llama-3.2-3B-Ins2026.02 | 69.5 | |
| Base ModelModel=Qwen2.5-7B-Ins2026.02 | 69.1 | |
| Base ModelModel=Llama-3.2-3B-Ins2026.02 | 30.7 |