Safety Evaluation on Llama-Guard 3-8B
2.38ASRMLP
Evaluation Results
| Method | Links | |
|---|---|---|
| MLPBase Model=Qwen2.5-72B, Condition=Cured2026.05 | 2.38 | |
| MLPBase Model=Llama-3.1-70B, Condition=Cured2026.05 | 4 | |
| MLPBase Model=Qwen2.5-7B, Condition=Cured2026.05 | 4.62 | |
| MoEBase Model=Gemma-2-27B, Condition=Cured2026.05 | 5.07 | |
| MoEBase Model=Gemma-2-9b, Condition=Cured2026.05 | 5.75 | |
| MoEBase Model=Llama-3.1-8B, Condition=Cured2026.05 | 7.19 | |
| MLPBase Model=Gemma-2-27B, Condition=Cured2026.05 | 9.12 | |
| MoEBase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 9.31 | |
| FMBase Model=Gemma-2-9b, Condition=Cured2026.05 | 11.94 | |
| FMBase Model=Gemma-2-27B, Condition=Cured2026.05 | 12.69 | |
| MLPBase Model=Gemma-2-9b, Condition=Cured2026.05 | 16.94 | |
| MoEBase Model=Llama-3.1-70B, Condition=Cured2026.05 | 17 | |
| MLPBase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 18.38 | |
| FMBase Model=Llama-3.1-8B, Condition=Cured2026.05 | 18.69 | |
| CVAEBase Model=Qwen2.5-72B, Condition=Cured2026.05 | 18.86 | |
| MoEBase Model=Qwen2.5-72B, Condition=Cured2026.05 | 19.06 | |
| MoEBase Model=Qwen2.5-7B, Condition=Cured2026.05 | 22.2 | |
| MLPBase Model=Llama-3.1-8B, Condition=Cured2026.05 | 24.12 | |
| CVAEBase Model=Gemma-2-9b, Condition=Cured2026.05 | 30.25 | |
| Safe LoRABase Model=Qwen2.5-72B, Condition=Cured2026.05 | 31.93 | |
| FMBase Model=Qwen2.5-7B, Condition=Cured2026.05 | 34.11 | |
| FMBase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 36.56 | |
| AEBase Model=Gemma-2-27B, Condition=Cured2026.05 | 36.65 | |
| CVAEBase Model=Qwen2.5-7B, Condition=Cured2026.05 | 36.74 | |
| Safe LoRABase Model=Gemma-2-9b, Condition=Cured2026.05 | 39.06 | |
| FMBase Model=Qwen2.5-72B, Condition=Cured2026.05 | 39.38 | |
| CVAEBase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 40.75 | |
| AEBase Model=Gemma-2-9b, Condition=Cured2026.05 | 41.21 | |
| FMBase Model=Llama-3.1-70B, Condition=Cured2026.05 | 41.75 | |
| CVAEBase Model=Llama-3.1-8B, Condition=Cured2026.05 | 42.12 | |
| Safe LoRABase Model=Llama-3.1-8B, Condition=Cured2026.05 | 43.75 | |
| Safe LoRABase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 44 | |
| CVAEBase Model=Gemma-2-27B, Condition=Cured2026.05 | 44.97 | |
| AEBase Model=Qwen2.5-72B, Condition=Cured2026.05 | 48.25 | |
| Safe LoRABase Model=Qwen2.5-7B, Condition=Cured2026.05 | 49.75 | |
| AEBase Model=Llama-3.1-8B, Condition=Cured2026.05 | 51.76 | |
| AEBase Model=Qwen2.5-7B, Condition=Cured2026.05 | 53.04 | |
| CVAEBase Model=Llama-3.1-70B, Condition=Cured2026.05 | 53.37 | |
| Safe LoRABase Model=Gemma-2-27B, Condition=Cured2026.05 | 55.62 | |
| Safe LoRABase Model=Llama-3.1-70B, Condition=Cured2026.05 | 56.38 | |
| AEBase Model=Llama-3.1-70B, Condition=Cured2026.05 | 58.88 | |
| AEBase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 59.88 | |
| Unsafe BaselineBase Model=Qwen2.5-72B, Condition=Unsafe (Finetuned)2026.05 | 60.25 | |
| SaLoRABase Model=Qwen2.5-72B, Condition=Cured2026.05 | 60.75 | |
| Unsafe BaselineBase Model=Gemma-2-9b, Condition=Unsafe (Finetuned)2026.05 | 63.04 | |
| SaLoRABase Model=Gemma-2-9b, Condition=Cured2026.05 | 63.23 | |
| Unsafe BaselineBase Model=Llama-3.1-70B, Condition=Unsafe (Finetuned)2026.05 | 65.81 | |
| SaLoRABase Model=Llama-3.1-70B, Condition=Cured2026.05 | 65.94 | |
| Unsafe BaselineBase Model=Gemma-2-27B, Condition=Unsafe (Finetuned)2026.05 | 66.37 | |
| Unsafe BaselineBase Model=Llama-3.1-8B, Condition=Unsafe (Finetuned)2026.05 | 66.85 | |
| SaLoRABase Model=Gemma-2-27B, Condition=Cured2026.05 | 67.29 | |
| SaLoRABase Model=Llama-3.1-8B, Condition=Cured2026.05 | 67.46 | |
| Unsafe BaselineBase Model=Qwen2.5-7B, Condition=Unsafe (Finetuned)2026.05 | 73.36 | |
| SaLoRABase Model=Qwen2.5-7B, Condition=Cured2026.05 | 73.47 | |
| Unsafe BaselineBase Model=Mistral-7B-v0.3, Condition=Unsafe (Finetuned)2026.05 | 74.25 | |
| SaLoRABase Model=Mistral-7B-v0.3, Condition=Cured2026.05 | 74.94 |