Machine Unlearning on TOFU 5% forget
0.1434LossPALU
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PALUBase Model=Llama-3.1-8B2026.01 | 0.1434 | 0.2379 | 0.1237 | 0.14 | |
| NPOBase Model=Llama-2-7B2026.01 | 0.1734 | 0.1998 | 0.174 | 0.1599 | |
| PALUBase Model=Llama-2-7B2026.01 | 0.184 | 0.1296 | 0.1643 | 0.1955 | |
| DPOBase Model=Llama-3.1-8B2026.01 | 0.2265 | 0.2262 | 0.1756 | 0.2603 | |
| SimNPOBase Model=Llama-3.1-8B2026.01 | 0.2393 | 0.202 | 0.2469 | 0.6362 | |
| TPOBase Model=Llama-2-7B2026.01 | 0.2459 | 0.1944 | 0.3857 | 0.7795 | |
| TPOBase Model=Llama-3.1-8B2026.01 | 0.3147 | 0.2796 | 0.2954 | 0.2269 | |
| SimNPOBase Model=Llama-2-7B2026.01 | 0.3176 | 0.293 | 0.3065 | 0.352 | |
| RetainBase Model=Llama-2-7B2026.01 | 0.3568 | 0.3106 | 0.3513 | 0.4641 | |
| DPOBase Model=Llama-2-7B2026.01 | 0.3614 | 0.276 | 0.3658 | 0.8059 | |
| RetainBase Model=Llama-3.1-8B2026.01 | 0.362 | 0.2958 | 0.3562 | 0.488 | |
| PDUBase Model=Llama-3.1-8B2026.01 | 0.4378 | 0.3889 | 0.4474 | 0.9103 | |
| PDUBase Model=Llama-2-7B2026.01 | 0.4729 | 0.5563 | 0.709 | 0.8879 | |
| NPOBase Model=Llama-3.1-8B2026.01 | 0.5777 | 0.511 | 0.569 | 0.5162 | |
| GDBase Model=Llama-3.1-8B2026.01 | 0.623 | 0.5591 | 0.6064 | 0.4316 | |
| GABase Model=Llama-2-7B2026.01 | 0.9243 | 0.8202 | 0.9603 | 0.7103 | |
| GDBase Model=Llama-2-7B2026.01 | 0.9243 | 0.8202 | 0.9603 | 0.7103 | |
| GABase Model=Llama-3.1-8B2026.01 | 0.9422 | 0.9277 | 0.9434 | 0.8517 | |
| OriginalBase Model=Llama-2-7B2026.01 | 1 | 1 | 1 | 1 | |
| OriginalBase Model=Llama-3.1-8B2026.01 | 1 | 1 | 1 | 0.9999 |