ResearchBenchmarksPreference Ranking on UltraFeedback Hard Ambiguous (test)Follow70Ranking AccuracySA-DPO58.5661.5364.567.47Apr 30, 2026Evaluation ResultsMethodMethodLinksRanking AccuracySA-DPOBackbone=Llama-3-8B, F...Backbone=Llama-3-8B, Fine-tuning Protocol=LoRA2026.0470SimPOBackbone=Llama-3-8B, F...Backbone=Llama-3-8B, Fine-tuning Protocol=LoRA2026.0465DPOBackbone=Llama-3-8B, F...Backbone=Llama-3-8B, Fine-tuning Protocol=LoRA2026.0459