Machine Unlearning on Unlearning Evaluation Dataset Forget, Retain, and Non-member sets
83.3SQSNSPU
Evaluation results
| Method | Links | ||
|---|---|---|---|
| NSPUBackbone=Llama-7B2026.01 | 83.3 | — | |
| GABackbone=Llama-7B2026.01 | 79.8 | — | |
| KLMBackbone=Llama-7B2026.01 | 79.8 | — | |
| NSPUBackbone=Mistral-7B2026.01 | 79.4 | — | |
| GABackbone=Mistral-7B2026.01 | 78.5 | — | |
| KLMBackbone=Mistral-7B2026.01 | 78.5 | — | |
| GABackbone=OLMoE-1B-7B2026.01 | 76.5 | — | |
| KLMBackbone=OLMoE-1B-7B2026.01 | 76.5 | — | |
| GDBackbone=Llama-7B2026.01 | 75.1 | — | |
| NSPUBackbone=OLMoE-1B-7B2026.01 | 74.2 | — | |
| NPOBackbone=Mistral-7B2026.01 | 73.7 | — | |
| NSPUBackbone=Llama-13B2026.01 | 72.3 | — | |
| DPOBackbone=OLMoE-1B-7B2026.01 | 70.6 | — | |
| DPOBackbone=Llama-7B2026.01 | 69.8 | — | |
| NPOBackbone=Llama-13B2026.01 | 69.4 | — | |
| GDBackbone=OLMoE-1B-7B2026.01 | 69.1 | — | |
| NPOBackbone=Llama-7B2026.01 | 65 | — | |
| GDBackbone=Mistral-7B2026.01 | 62.9 | — | |
| GDBackbone=Llama-13B2026.01 | 56.5 | — | |
| DPOBackbone=Llama-13B2026.01 | 52.8 | — | |
| DPOBackbone=Mistral-7B2026.01 | 46.2 | — | |
| GABackbone=Llama-13B2026.01 | 40.3 | — | |
| KLMBackbone=Llama-13B2026.01 | 40.3 | — | |
| NPOBackbone=OLMoE-1B-7B2026.01 | 17.5 | — | |
| DPOBackbone=Average2026.01 | — | 0.599 | |
| GABackbone=Average2026.01 | — | 0.688 | |
| GDBackbone=Average2026.01 | — | 0.659 | |
| KLMBackbone=Average2026.01 | — | 0.688 | |
| NPOBackbone=Average2026.01 | — | 0.564 | |
| NSPUBackbone=Average2026.01 | — | 0.773 |