Machine Unlearning on Internal e-commerce benchmark small-scale seller, 66 items
86.4Forget Set ROUGEME+GD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ME+GDBackbone=Llama 3.1 8B2025.05 | 86.4 | 0.72 | 72.5 | 0.52 | |
| BaselineBackbone=Llama 3.1 8B2025.05 | 85.6 | 0.72 | 72 | 0.52 | |
| Undial+KLBackbone=Llama 3.1 8B2025.05 | 64.3 | 0.9 | 69 | 0.55 | |
| Unilogit+KLBackbone=Llama 3.1 8B, variant=32025.05 | 52.7 | 1.23 | 71.6 | 0.54 | |
| NPO+KLBackbone=Llama 3.1 8B, variant=22025.05 | 50.6 | 1.86 | 67.1 | 0.74 | |
| NPO+KLBackbone=Llama 3.1 8B, variant=12025.05 | 39.6 | 1.81 | 58.8 | 0.75 | |
| Unilogit+KLBackbone=Llama 3.1 8B, variant=22025.05 | 10.2 | 6.84 | 62.5 | 0.82 | |
| Unilogit+KLBackbone=Llama 3.1 8B, variant=12025.05 | 7.7 | 6.16 | 51.6 | 0.9 |