Machine Unlearning on TOFU Forget01 (1%) benchmark leaderboard · SOTA2 Research

Machine Unlearning on TOFU Forget01 (1%)

96Forget Quality (Rouge-L)DPO

Evaluation Results

Method	Links
DPO2025.09		96	49	0.0089	92	48	43	45	76	50	52
KL2025.09		96	48	0.0074	92	48	43	45	76	50	52
DPO2025.09		96	48	0.0089	92	48	44	45	75	50	52
NPO2025.09		96	48	0.0061	92	48	43	45	76	50	52
GA2025.09		96	48	0.0061	92	48	43	45	76	50	52
GD2025.09		96	48	0.0074	92	48	42	45	76	50	52
IHL2025.09		96	48	0.0042	92	48	43	45	75	50	52
IHL2025.09		94	48	0.0074	92	48	43	45	75	50	52
Original2025.09		93	48	0	92	48	41	45	75	50	52
GD2025.09		93	48	0.0074	92	48	43	45	76	50	52
Original2025.09		93	48	0	92	48	41	45	75	50	52
NPO2025.09		92	48	0.0061	91	48	43	45	76	50	52
GA2025.09		92	48	0.0042	92	48	43	45	77	50	52
KL2025.09		91	48	0.0061	92	48	43	45	77	50	52
LoKU2025.09		50	49	0.0128	83	49	37	45	73	50	51
LoKU2025.09		47	51	0.0337	80	49	34	46	73	51	50
Retain902025.09		43	63	100	91	48	43	45	76	49	52
GD+Sine2025.09		38	48	94.3	93	48	41	46	77	50	52
GD+Sine2025.09		35	48	94.3	93	48	41	46	77	49	52
Retain902025.09		33	63	100	91	48	43	45	76	49	52
GA+FILA2025.09		4	76	0.107	6	21	1	29	2	30	0
GD+FILA2025.09		3	69	3.24	6	20	0	31	3	31	0
GD+FILA2025.09		1	65	4.55	1	24	0	32	2	36	0
GA+FILA2025.09		0	65	2.72	1	22	0	32	0	34	0