Unlearning on Toxic Bias Dataset
90.12Task AccuracyFT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FTModel=Llama-3.1-8B2026.07 | 90.12 | 94.33 | 61.7 | 3 | |
| MPTuneModel=Llama-3.1-8B2026.07 | 90.06 | 64.12 | 62.1 | 2.3 | |
| MPSelectTuneModel=Llama-3.1-8B2026.07 | 89.93 | 58.34 | 62.8 | 1.6 | |
| FTModel=Llama-2-7B2026.07 | 89.92 | 95.67 | 41.1 | 5 | |
| MPSelectTuneModel=Llama-2-7B2026.07 | 89.75 | 53.13 | 42 | 2.6 | |
| MPTuneModel=Llama-2-7B2026.07 | 89.63 | 60.17 | 41.9 | 2.8 | |
| FTModel=Mistral-7B-Instruct-v0.32026.07 | 88.7 | 94.8 | 53.4 | 2.5 | |
| MPSelectTuneModel=Mistral-7B-Instruct-v0.32026.07 | 88.6 | 56.8 | 53.4 | 2.1 | |
| MPTuneModel=Mistral-7B-Instruct-v0.32026.07 | 88.4 | 62.5 | 53.2 | 2.3 | |
| ICULModel=Llama-2-7B2026.07 | 86.5 | 66.96 | 41.1 | 5.6 | |
| ICULModel=Mistral-7B-Instruct-v0.32026.07 | 85.2 | 68.3 | 53.4 | 10.1 | |
| AugModel=Llama-3.1-8B2026.07 | 84.36 | 75.17 | 58.3 | 11.9 | |
| AugModel=Mistral-7B-Instruct-v0.32026.07 | 82.1 | 84.9 | 52.8 | 6.7 | |
| AugModel=Llama-2-7B2026.07 | 81.46 | 86.33 | 39.4 | 13.5 | |
| ICULModel=Llama-3.1-8B2026.07 | 81.35 | 65.97 | 61.7 | 13.4 | |
| SKUModel=Llama-3.1-8B2026.07 | 80.63 | 69.42 | 60.3 | 15.6 | |
| SKUModel=Llama-2-7B2026.07 | 80.46 | 68.33 | 38.6 | 11.4 | |
| SKUModel=Mistral-7B-Instruct-v0.32026.07 | 79.8 | 70.1 | 50.3 | 15.6 | |
| BaseModel=Llama-3.1-8B2026.07 | 77.66 | 83.41 | 65 | 16.6 | |
| BaseModel=Mistral-7B-Instruct-v0.32026.07 | 76.3 | 83.1 | 56.1 | 12.9 | |
| BaseModel=Llama-2-7B2026.07 | 75.41 | 82.25 | 43.9 | 11.6 |