Machine Unlearning on MUSE forget-set (Quantization Metrics)
0.365Performance (BF16)Zero-shot
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Zero-shotBackbone=Llama-3.1-8B-Instruct2026.05 | 0.365 | — | — | |
| LUNARBackbone=Llama-3.1-8B-Instruct2026.05 | 0.187 | 0.198 | 0.011 | |
| LUNARBackbone=Qwen-3-8B2026.05 | 0.162 | 0.171 | 0.009 | |
| Zero-shotBackbone=Qwen-3-8B2026.05 | 0.024 | — | — | |
| MANSUBackbone=Qwen-3-8B2026.05 | 0.021 | 0.017 | -0.004 | |
| Surgical GABackbone=Qwen-3-8B2026.05 | 0.02 | 0.017 | -0.003 | |
| NPOBackbone=Llama-3.1-8B-Instruct2026.05 | 0.013 | 0.016 | 0.002 | |
| GU+SimNPOBackbone=Qwen-3-8B2026.05 | 0.013 | 0.017 | 0.004 | |
| NPOBackbone=Qwen-3-8B2026.05 | 0.011 | 0.021 | 0.009 | |
| MANSUBackbone=Llama-3.1-8B-Instruct2026.05 | 0.005 | 0.003 | -0.002 | |
| SimNPOBackbone=Qwen-3-8B2026.05 | 0.001 | 0.013 | 0.012 | |
| Global GABackbone=Llama-3.1-8B-Instruct2026.05 | 0 | 0 | 0 | |
| Surgical GABackbone=Llama-3.1-8B-Instruct2026.05 | 0 | 0 | 0 | |
| SimNPOBackbone=Llama-3.1-8B-Instruct2026.05 | 0 | 0 | 0 | |
| GU+SimNPOBackbone=Llama-3.1-8B-Instruct2026.05 | 0 | 0 | 0 | |
| Global GABackbone=Qwen-3-8B2026.05 | 0 | 0.009 | 0.009 |