LLM Unlearning on UNCD-Cyber (Retain)
57.19In-Domain AccuracyBase Model
Evaluation results
| Method | Links | |
|---|---|---|
| Base ModelBackbone=Llama-3-8B2025.02 | 57.19 | |
| RMUBackbone=Llama-3-8B2025.02 | 56.55 | |
| NPOGDRBackbone=Llama-3-8B2025.02 | 55.27 | |
| Base ModelBackbone=Mistral-7B2025.02 | 54.21 | |
| NPOKLRBackbone=Llama-3-8B2025.02 | 52.34 | |
| GAKLRBackbone=Llama-3-8B2025.02 | 52.13 | |
| TVBackbone=Llama-3-8B2025.02 | 49.57 | |
| RMUBackbone=Mistral-7B2025.02 | 49.57 | |
| GAKLRBackbone=Mistral-7B2025.02 | 48.61 | |
| NPOKLRBackbone=Mistral-7B2025.02 | 48.19 | |
| NPOGDRBackbone=Mistral-7B2025.02 | 39.66 | |
| NPOBackbone=Llama-3-8B2025.02 | 33.37 | |
| GAGDRBackbone=Llama-3-8B2025.02 | 30.17 | |
| TVBackbone=Mistral-7B2025.02 | 27.99 | |
| GAGDRBackbone=Mistral-7B2025.02 | 18.76 | |
| NPOBackbone=Mistral-7B2025.02 | 17.53 | |
| GABackbone=Llama-3-8B2025.02 | 16 | |
| GABackbone=Mistral-7B2025.02 | 15.83 |