Safety Evaluation on HH-RLHF (test)
1.02Harm ScoreSAILS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SAILSModel=Gemma-2-9B, # Params=0.19%2025.12 | 1.02 | 0.996 | 0.4 | |
| SAILSModel=Llama-3.1-8B, # Params=0.21%2025.12 | 1.03 | 0.992 | 0.8 | |
| IT+RLModel=Gemma-2-9B, # Params=100%2025.12 | 1.08 | 0.982 | 0 | |
| IT+RLModel=Llama-3.1-8B, # Params=100%2025.12 | 1.13 | 0.978 | 2.6 | |
| SAILSModel=Gemma-2-2B, # Params=0.24%2025.12 | 1.17 | 0.968 | 2.6 | |
| IT+RLModel=Gemma-2-2B, # Params=100%2025.12 | 1.18 | 0.946 | 2.6 | |
| PromptModel=Gemma-2-9B, # Params=-2025.12 | 1.29 | 0.942 | 4.6 | |
| FFTModel=Llama-3.1-8B, # Params=100%2025.12 | 1.34 | 0.928 | 6 | |
| LoRAModel=Llama-3.1-8B, # Params=0.21%2025.12 | 1.4 | 0.92 | 5.2 | |
| LoRAModel=Gemma-2-9B, # Params=0.19%2025.12 | 1.41 | 0.924 | 4.8 | |
| FFTModel=Gemma-2-9B, # Params=100%2025.12 | 1.44 | 0.922 | 7 | |
| PromptModel=Llama-3.1-8B, # Params=-2025.12 | 1.47 | 0.906 | 7.6 | |
| PromptModel=Gemma-2-2B, # Params=-2025.12 | 1.51 | 0.88 | 8.8 | |
| FFTModel=Gemma-2-2B, # Params=100%2025.12 | 1.52 | 0.892 | 8.8 | |
| DoRAModel=Gemma-2-2B, # Params=0.25%2025.12 | 1.54 | 0.89 | 7.6 | |
| DoRAModel=Llama-3.1-8B, # Params=0.22%2025.12 | 1.54 | 0.896 | 7.6 | |
| LoRAModel=Gemma-2-2B, # Params=0.24%2025.12 | 1.56 | 0.876 | 8.4 | |
| DoRAModel=Gemma-2-9B, # Params=0.20%2025.12 | 1.67 | 0.86 | 10.2 | |
| OriginalModel=Llama-3.1-8B, # Params=-2025.12 | 2.54 | 0.62 | 35.2 | |
| OriginalModel=Gemma-2-9B, # Params=-2025.12 | 2.73 | 0.576 | 37.4 | |
| OriginalModel=Gemma-2-2B, # Params=-2025.12 | 2.88 | 0.528 | 42.4 |