Toxicity Evaluation on AttaQ
0.04Max Toxicity ScoreRAD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RADBackbone=Qwen3-4B, Beta (β)=500, Temperature=0.1, Trials=32026.01 | 0.04 | 0.033 | 13.22 | 0.1 | |
| RADBackbone=Qwen3-4B, Beta (β)=300, Temperature=0.1, Trials=32026.01 | 0.053 | 0.041 | 5.78 | 0.2 | |
| RADBackbone=Qwen3-4B, Beta (β)=100, Temperature=0.1, Trials=32026.01 | 0.07 | 0.055 | 2.79 | 0.1 | |
| RADBackbone=Qwen3-4B, Beta (β)=75, Temperature=0.1, Trials=32026.01 | 0.073 | 0.058 | 2.65 | 0.2 | |
| TIDEBackbone=Qwen3-4B, Temperature=0.1, Trials=3, # Iterations K=1.022026.01 | 0.076 | 0.063 | 2.47 | 0 | |
| RADBackbone=Qwen3-4B, Beta (β)=50, Temperature=0.1, Trials=32026.01 | 0.077 | 0.061 | 2.56 | 0.4 | |
| RADBackbone=Qwen3-4B, Beta (β)=10, Temperature=0.1, Trials=32026.01 | 0.091 | 0.072 | 2.46 | 1.1 | |
| SASABackbone=Qwen3-4B, Beta (β)=500, Temperature=0.1, Trials=32026.01 | 0.096 | 0.074 | 3.17 | 1.1 | |
| SASABackbone=Qwen3-4B, Beta (β)=300, Temperature=0.1, Trials=32026.01 | 0.097 | 0.076 | 2.61 | 1.1 | |
| Base modelBackbone=Qwen3-4B, Temperature=0.1, Trials=32026.01 | 0.098 | 0.076 | 2.46 | 1.6 | |
| SASABackbone=Qwen3-4B, Beta (β)=100, Temperature=0.1, Trials=32026.01 | 0.098 | 0.076 | 2.83 | 1.6 | |
| SASABackbone=Qwen3-4B, Beta (β)=50, Temperature=0.1, Trials=32026.01 | 0.1 | 0.077 | 2.83 | 1.4 | |
| SASABackbone=Qwen3-4B, Beta (β)=75, Temperature=0.1, Trials=32026.01 | 0.1 | 0.077 | 2.84 | 1.4 | |
| SASABackbone=Qwen3-4B, Beta (β)=10, Temperature=0.1, Trials=32026.01 | 0.103 | 0.078 | 2.44 | 1.6 |