Preference Alignment on Psoups (test) benchmark leaderboard · SOTA2 Research

Preference Alignment on Psoups (test)

1.39Helpfulness (RM)MetaAligner

Evaluation Results

Method	Links
MetaAligner2025.04		1.39	0.55	66	0.89	0.54	74	0.84
Aligner2025.04		1.32	0.67	72	0.97	0.63	70	0.9
Persona-judge2025.04		1.29	0.63	84	0.98	0.82	98	0.93
Persona-judge-Base2025.04		1.21	0.59	80	1.01	0.79	88	0.9
Steering2025.04		1.17	0.63	38	0.96	0.81	66	0.89
Args2025.04		1.09	—	74	—	0.98	94	0.9
Base2025.04		1.06	0.61	—	0.97	0.83	—	0.87
MOD2025.04		0.93	0.55	60	0.96	0.92	84	0.84
MORLHF2025.04		0.91	0.31	14	0.88	0.84	4	0.73
RiC2025.04		0.9	0.54	40	0.97	0.9	70	0.83
MODPO2025.04		0.89	0.56	52	0.96	0.77	80	0.8
Rewarded soups2025.04		0.87	0.5	34	0.95	0.87	64	0.8
Personalized soups2025.04		-0.72	0.38	72	0.92	0.73	92	0.53