ResearchBenchmarksHelpful Assistant Evaluation (Helpfulness and Harmlessness) on HH-RLHFFollow1.98Helpfulness Value (HV)MI-EPO0.39920.80961.221.6304Jul 1, 2026Evaluation ResultsMethodMethodLinksHelpfulness Value (HV)Model Improvement (MIP)CRD ScoreMI-EPOBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.071.98680.52MO-ODPOBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.071.68640.57RSBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.070.68210.58RiCBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.070.4671.05