Toxicity Reduction on Toxicity Evaluation Prompts
0.4LG Score (Before)TRNE
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| TRNEModel=Gemma, Component=MLP, Layers=5, Edit Strength=22026.05 | 0.4 | 0 | 0.4 | 28.2 | 24.8 | 3.4 | 29.91 | 30.65 | |
| TRNEModel=Llama3_2, Component=MLP, Layers=5, Edit Strength=52026.05 | 2.2 | 1 | 1.2 | 37.8 | 32.2 | 5.6 | 22.17 | 24.83 | |
| TRNEModel=Llama3_2, Component=MLP, Layers=10, Edit Strength=202026.05 | 2.2 | 0 | 2.2 | 39.8 | 0 | 39.8 | 22.17 | 9,297.41 | |
| TRNEModel=Qwen-2, Component=MLP, Layers=10, Edit Strength=52026.05 | 2.6 | 2.6 | 0 | 44.8 | 38.2 | 6.6 | 18.19 | 35.8 | |
| TRNEModel=Tiny-llama, Component=MLP, Layers=10, Edit Strength=22026.05 | 4.4 | 3 | 1.4 | 44.6 | 29.8 | 14.8 | 17.59 | 19.45 |