Toxicity Mitigation on Toxicity prompts
0.12CLS Toxicity (%)A-LQR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| A-LQRModel=Llama-3-8B2026.04 | 0.12 | 0.65 | 8.35 | |
| A-LQRModel=Qwen-2.5-14B2026.04 | 0.12 | 0.7 | 9.01 | |
| ActAddModel=Qwen-2.5-14B2026.04 | 0.16 | 0.69 | 10.81 | |
| A-LQRModel=Gemma-2-2B2026.04 | 0.18 | 0.68 | 12.26 | |
| A-LQRModel=Qwen2.5-32B2026.04 | 0.22 | 0.68 | 8.83 | |
| S-PIDModel=Qwen2.5-32B2026.04 | 0.28 | 0.71 | 8.86 | |
| Mean-AcTModel=Gemma-2-2B2026.04 | 0.5 | 0.68 | 8.7 | |
| ActAddModel=Llama-3-8B2026.04 | 0.56 | 0.63 | 12.56 | |
| ODESteerModel=Gemma-2-2B2026.04 | 0.58 | 0.64 | 11.62 | |
| S-PIDModel=Qwen-2.5-14B2026.04 | 0.6 | 0.71 | 8.17 | |
| ITIModel=Llama-3-8B2026.04 | 0.64 | 0.67 | 9.39 | |
| S-PIDModel=Llama-3-8B2026.04 | 0.66 | 0.66 | 8.1 | |
| ODESteerModel=Qwen-2.5-14B2026.04 | 0.68 | 0.65 | 10.82 | |
| PID-AcTModel=Llama-3-8B2026.04 | 0.7 | 0.65 | 7.59 | |
| ODESteerModel=Llama-3-8B2026.04 | 0.7 | 0.66 | 8.72 | |
| Mean-AcTModel=Llama-3-8B2026.04 | 0.72 | 0.63 | 7.92 | |
| Linear-AcTModel=Qwen-2.5-14B2026.04 | 0.78 | 0.68 | 8.94 | |
| S-PIDModel=Gemma-2-2B2026.04 | 0.8 | 0.7 | 11.68 | |
| PID-AcTModel=Gemma-2-2B2026.04 | 0.86 | 0.69 | 8.2 | |
| Linear-AcTModel=Gemma-2-2B2026.04 | 0.92 | 0.69 | 8.77 | |
| ITIModel=Gemma-2-2B2026.04 | 0.96 | 0.67 | 10.24 | |
| ITIModel=Qwen-2.5-14B2026.04 | 0.96 | 0.76 | 8.79 | |
| Linear-AcTModel=Llama-3-8B2026.04 | 1.04 | 0.68 | 7.46 | |
| ActAddModel=Gemma-2-2B2026.04 | 1.1 | 0.64 | 11.42 | |
| ITIModel=Qwen2.5-32B2026.04 | 1.62 | 0.73 | — | |
| Linear-AcTModel=Qwen2.5-32B2026.04 | 3.14 | 0.7 | 8.41 | |
| OriginalModel=Qwen-2.5-14B2026.04 | 3.26 | 0.7 | 8.3 | |
| PID-AcTModel=Qwen-2.5-14B2026.04 | 3.52 | 0.69 | 8.99 | |
| Mean-AcTModel=Qwen-2.5-14B2026.04 | 3.64 | 0.67 | 10.03 | |
| OriginalModel=Qwen2.5-32B2026.04 | 3.72 | 0.7 | 8.61 | |
| OriginalModel=Gemma-2-2B2026.04 | 4.16 | 0.67 | 8.95 | |
| OriginalModel=Llama-3-8B2026.04 | 5.14 | 0.67 | 6.74 |