Machine Unlearning on TOFU Llama 3.1 8B (10%) benchmark leaderboard · SOTA2 Research

Machine Unlearning on TOFU Llama 3.1 8B (10%)

0.671Forgetting QualityT3-Unlearning

Evaluation Results

Method	Links
T3-Unlearning2026.02		0.671	0.899	0.612	42.3	0.992	0.462	54.3	0.863	0.686	50.6	0.856	0.641
SimNPO2026.02		0.536	0.84	0.662	87	0.745	0.529	50.6	0.904	0.685	50.2	0.89	0.649
SatImp2026.02		0.219	0.301	0.375	52.4	0.531	0.444	36.8	0.517	0.494	35.3	0.165	0.462
NPO2026.02		0.185	0.521	0.493	60.2	0.542	0.468	42.2	0.664	0.569	39.3	0.416	0.506
GradDiff2026.02		0.065	0	0	1.8	0.035	0.364	23.9	0.001	0.335	21.6	0	0.359
WGA2026.02		0.02	0.834	0.629	87.3	0.748	0.502	51.1	0.898	0.691	43	0.875	0.545
IdkDPO2026.02		0.02	0.413	0.465	65.6	0.509	0.486	43.1	0.655	0.586	40.2	0.271	0.522
ULD2026.02		0.012	0.944	0.642	98.8	0.978	0.533	49.4	0.904	0.638	41.2	0.952	0.517
RMU2026.02		0.001	0	0	0	0.072	0.299	22.9	0.023	0.588	29.8	0	0.477
GradAscent2026.02		0	0.95	0.638	99.1	0.995	0.531	48.1	0.91	0.63	40.9	0.949	0.512
UNDIAL2026.02		0	0.927	0.697	97.1	0.961	0.517	56.1	0.908	0.729	50.4	0.914	0.633