Detoxification on BOLD
1.9Toxicity (Max)RAD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=5002026.01 | 1.9 | 1.6 | 11.93 | 0 | |
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=3002026.01 | 2 | 1.6 | 7.91 | 0 | |
| RADBackbone=Gemma 2-2B, beta=3002026.01 | 2.5 | 2 | 6.54 | 0 | |
| RADBackbone=Gemma 2-2B, beta=5002026.01 | 2.6 | 2.2 | 10.4 | 0 | |
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=752026.01 | 2.7 | 2.1 | 4.16 | 0 | |
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=1002026.01 | 2.7 | 2.1 | 4.44 | 0 | |
| RADBackbone=Gemma 2-2B, beta=1002026.01 | 2.8 | 2.2 | 3.59 | 0 | |
| RADBackbone=Gemma 2-2B, beta=752026.01 | 3 | 2.4 | 3.38 | 0.1 | |
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=502026.01 | 3 | 2.3 | 3.98 | 0 | |
| RADBackbone=Gemma 2-2B, beta=502026.01 | 3.2 | 2.5 | 3.24 | 0 | |
| TIDEBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, # Iterations K=12026.01 | 3.2 | 2.6 | 3.77 | 0 | |
| TIDEBackbone=Gemma 2-2B, # Iterations K=12026.01 | 3.5 | 2.9 | 3.14 | 0 | |
| RADBackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=102026.01 | 3.7 | 2.8 | 3.78 | 0 | |
| SASABackbone=Gemma 2-2B, beta=5002026.01 | 3.8 | 2.8 | 9.53 | 0 | |
| RADBackbone=Gemma 2-2B, beta=102026.01 | 4 | 3 | 3.15 | 0 | |
| SASABackbone=Gemma 2-2B, beta=3002026.01 | 4.1 | 3 | 5.22 | 0.2 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=502026.01 | 4.2 | 3.2 | 3.81 | 0.1 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=752026.01 | 4.2 | 3.2 | 3.81 | 0.2 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=1002026.01 | 4.2 | 3.2 | 3.87 | 0.2 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=102026.01 | 4.3 | 3.2 | 3.81 | 0.2 | |
| Base modelBackbone=Llama 3.1-8B, Temperature=0.1, Trials=32026.01 | 4.4 | 3.2 | 3.79 | 0.3 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=3002026.01 | 4.4 | 3.3 | 4.93 | 0.2 | |
| SASABackbone=Gemma 2-2B, beta=502026.01 | 4.5 | 3.5 | 3.2 | 0.2 | |
| SASABackbone=Gemma 2-2B, beta=752026.01 | 4.5 | 3.4 | 3.24 | 0.2 | |
| SASABackbone=Gemma 2-2B, beta=1002026.01 | 4.5 | 3.4 | 3.31 | 0.3 | |
| Base modelBackbone=Gemma 2-2B2026.01 | 4.6 | 3.5 | 3.14 | 0.2 | |
| SASABackbone=Gemma 2-2B, beta=102026.01 | 4.6 | 3.5 | 3.17 | 0.2 | |
| SASABackbone=Llama 3.1-8B, Temperature=0.1, Trials=3, beta (β)=5002026.01 | 4.7 | 3.5 | 8.23 | 0.3 |