Unlearning on RT-Gender Dataset
73.58Task AccuracySKU
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SKUModel=Llama-3.1-8B2026.07 | 73.58 | 67.33 | 61.9 | 0.105 | |
| FTModel=Llama-3.1-8B2026.07 | 71.12 | 86.87 | 59.6 | 0.056 | |
| MPSelectTuneModel=Llama-3.1-8B2026.07 | 71.03 | 49.81 | 64.2 | 0.032 | |
| MPTuneModel=Llama-3.1-8B2026.07 | 70.96 | 54.33 | 64.4 | 0.029 | |
| FTModel=Llama-2-7B2026.07 | 70.08 | 86.42 | 40.2 | 0.043 | |
| MPSelectTuneModel=Llama-2-7B2026.07 | 70.08 | 51.5 | 43.1 | 0.011 | |
| MPTuneModel=Llama-2-7B2026.07 | 70 | 53.83 | 42.6 | 0.021 | |
| MPSelectTuneModel=Mistral-7B-Instruct-v0.32026.07 | 69.3 | 52.4 | 53.7 | 0.021 | |
| MPTuneModel=Mistral-7B-Instruct-v0.32026.07 | 69.1 | 55.7 | 53.8 | 0.023 | |
| FTModel=Mistral-7B-Instruct-v0.32026.07 | 68.9 | 85.3 | 53.8 | 0.043 | |
| ICULModel=Llama-2-7B2026.07 | 67.43 | 73.25 | 40.2 | 0.118 | |
| AugModel=Llama-3.1-8B2026.07 | 67.31 | 77.35 | 59.7 | 0.123 | |
| ICULModel=Mistral-7B-Instruct-v0.32026.07 | 66.8 | 71.6 | 53.8 | 0.118 | |
| AugModel=Mistral-7B-Instruct-v0.32026.07 | 65.4 | 80.7 | 54 | 0.108 | |
| SKUModel=Llama-2-7B2026.07 | 65.36 | 59.45 | 37.4 | 0.121 | |
| SKUModel=Mistral-7B-Instruct-v0.32026.07 | 64.7 | 65.2 | 52.3 | 0.121 | |
| ICULModel=Llama-3.1-8B2026.07 | 64.22 | 66.93 | 59.6 | 0.144 | |
| AugModel=Llama-2-7B2026.07 | 64.17 | 82.5 | 37.6 | 0.108 | |
| BaseModel=Llama-3.1-8B2026.07 | 63.39 | 75.36 | 65 | 0.173 | |
| BaseModel=Mistral-7B-Instruct-v0.32026.07 | 59.2 | 72.8 | 56.1 | 0.146 | |
| BaseModel=Llama-2-7B2026.07 | 58.54 | 71.3 | 43.9 | 0.146 |