LLM Unlearning on UNCD-Cyber (Forget)
61.96AccuracyBase Model
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Base ModelBackbone=Llama-3-8B2025.02 | 61.96 | 57.26 | 69.83 | 46 | |
| Base ModelBackbone=Mistral-7B2025.02 | 58.92 | 59.44 | 72.59 | 44 | |
| RMUBackbone=Llama-3-8B2025.02 | 58.68 | 55.43 | 67.43 | 36 | |
| NPOKLRBackbone=Llama-3-8B2025.02 | 57.39 | 48.76 | 65.97 | 15 | |
| TVBackbone=Llama-3-8B2025.02 | 56.47 | 53.98 | 68.7 | 27 | |
| GAKLRBackbone=Llama-3-8B2025.02 | 56.27 | 53.91 | 68.12 | 14 | |
| GAKLRBackbone=Mistral-7B2025.02 | 52.13 | 56.04 | 71.81 | 16 | |
| NPOKLRBackbone=Mistral-7B2025.02 | 51.77 | 56.62 | 71.9 | 17 | |
| NPOGDRBackbone=Llama-3-8B2025.02 | 50.1 | 48.02 | 67.24 | 13 | |
| RMUBackbone=Mistral-7B2025.02 | 48.86 | 49.17 | 69.07 | 37 | |
| NPOBackbone=Llama-3-8B2025.02 | 29.75 | 39.98 | 50.46 | -7 | |
| NPOGDRBackbone=Mistral-7B2025.02 | 27.24 | 44.1 | 45.14 | 14 | |
| TVBackbone=Mistral-7B2025.02 | 27.06 | 38.9 | 27.65 | 28 | |
| GAGDRBackbone=Mistral-7B2025.02 | 17.56 | 29.73 | 9.93 | 23 | |
| GAGDRBackbone=Llama-3-8B2025.02 | 16.81 | 21.05 | 12.25 | 21 | |
| GABackbone=Llama-3-8B2025.02 | 13.86 | 7.83 | 9.87 | -12 | |
| GABackbone=Mistral-7B2025.02 | 12.26 | 16.27 | 3.67 | -10 | |
| NPOBackbone=Mistral-7B2025.02 | 9.75 | 21.48 | 3.73 | -5 |