Selective capability unlearning on SpeechMassive Forget Set (DF) 1.0 benchmark leaderboard · SOTA2 Research

Selective capability unlearning on SpeechMassive Forget Set (DF) 1.0

81.55IFOriginal

Evaluation Results

Method	Links
Original2026.06		81.55	54.9	79.02	76.39
GA+GD2026.06		77.51	54.14	78.01	76.05
Original2026.06		77.38	52.03	77.02	74.7
GA+GD2026.06		74.06	50.17	67.55	67.99
GA+KL2026.06		72.2	48.68	74.05	74.7
GA+KL2026.06		67.3	48.01	68.22	71
NPO+KL2026.06		59.73	51.34	60.27	61.43
NPO+KL2026.06		55.14	49.76	56.87	58.43
RS2026.06		53.74	45.03	54.16	51.97
RS2026.06		51.07	47.5	57.39	59.47
NPO2026.06		50.19	50.11	49.83	49.17
GA2026.06		40.14	45.8	40.57	44.6
NPO2026.06		39.56	44.17	35.91	35.07
RL2026.06		34.76	41.27	30.46	32.11
GA2026.06		33.15	37.91	31.77	32.16
RL2026.06		28.63	36.42	24.94	26.18
BSU2026.06		25.37	27.48	23.64	26.17
BSU2026.06		22.41	24.33	29.87	29.94
Retrain2026.06		12.74	23.01	19.77	21.8
Retrain2026.06		10.95	20.14	22.57	21.87