Machine Unlearning on Harry Potter (test)
74.8BLEUOriginal
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OriginalBase Model=Llama-2-7B2026.03 | 74.8 | 85.1 | 74.1 | |
| WHPBase Model=Llama-2-7B2026.03 | 23.6 | 17.9 | 14.9 | |
| ELMBase Model=Llama-2-7B2026.03 | 8.1 | 9 | 2.7 | |
| LoRABase Model=Llama-2-7B2026.03 | 7.2 | 11.5 | 3.5 | |
| Fine-tuneBase Model=Llama-2-7B2026.03 | 6.4 | 17.2 | 5.9 | |
| A-LoRABase Model=Llama-2-7B2026.03 | 5.9 | 10.4 | 1.9 | |
| OursBase Model=Llama-2-7B2026.03 | 4.7 | 9.6 | 1.3 | |
| GDBase Model=Llama-2-7B2026.03 | 3.9 | 14.5 | 3.4 | |
| RetainBase Model=Llama-2-7B2026.03 | 1.9 | 9.8 | 0 | |
| NPOBase Model=Llama-2-7B2026.03 | 1.5 | 5.3 | 2.5 | |
| KLBase Model=Llama-2-7B2026.03 | 1.2 | 8.9 | 0.8 | |
| GABase Model=Llama-2-7B2026.03 | 0 | 0 | 6 |