Share your thoughts, 1 month free Claude Pro on usSee more

Model Editing on CounterFact (Efficacy, Generalization, Specificity, Edited Retention)

73.06Efficacy

AlphaEdit

Updated 1mo ago

Evaluation Results

Method	Links
AlphaEdit 2026.04		73.06	66.43	48.54	47.4
HiEditfull 2026.04		66.53	55.65	45.7	58.2
HiEditrand 2026.04		66.4	55.48	45.51	58
RLEdit 2026.04		66.35	55.26	44.79	57.2
FT 2026.04		65.24	57.99	44.03	62.8
HiEditfull 2026.04		64.06	52.8	43.59	54
FT 2026.04		60.13	39.49	52.26	42.7
HiEditrand 2026.04		59.81	50	47.37	50.3
ROME 2026.04		59.26	56.45	48.52	48.3
RLEdit 2026.04		55.51	49.92	46.28	45.8
AlphaEdit 2026.04		38.17	38.83	66.51	34
PRUNE 2026.04		32.41	29.78	65.1	29.8
HiEdit_full 2026.04		31.66	22.41	10.8	23.05
MEMIT 2026.04		31.51	28.78	65.58	30.1
HiEdit_rand 2026.04		31.45	22.35	10.96	22.75
RECT 2026.04		30.72	28.34	66.08	30
DAFNet 2026.04		29.56	31.55	68.38	33.6
RLEdit 2026.04		27.93	20.58	10.83	21
MEND 2026.04		27.28	28.54	70.2	27.7
MEMIT 2026.04		18.61	16.94	16.93	16.2
AlphaEdit 2026.04		11.64	4.87	1.49	1.6
MALMEN 2026.04		10.46	13.12	86.94	11.4
DAFNet 2026.04		10.23	12.74	86.6	13.5
Pre-edited 2026.04		8.23	10.8	88.72	10.4
MEND 2026.04		7.69	13.4	7.45	10.5
Pre-edited 2026.04		7.2	9.1	89.77	8.7
ROME 2026.04		1.78	1.8	1.17	0.9
Pre-edited 2026.04		1.01	1.01	21.74	1.35
FT 2026.04		0.29	0.06	0	0.03
PRUNE 2026.04		0	0	0	0
RECT 2026.04		0	0	0	0
MALMEN 2026.04		0	0	0	0
ROME 2026.04		0	0	0	0