ResearchBenchmarksClosed Loop RLHF on HH (40% noise)Follow57Win Rateα-URM48.1650.45552.7555.045Jul 3, 2026Evaluation ResultsMethodMethodLinksWin Rateα-URMBase Model=Qwen-3-1.7B...Base Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0757URMBase Model=Qwen-3-1.7B...Base Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0756.5RMBase Model=Qwen-3-1.7B...Base Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0748.5