List your company

Closed Loop RLHF on HH (40% noise)

57Win Rateα-URM
48.1650.45552.7555.045Jul 3, 2026

Evaluation results

MethodLinks
α-URMBase Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0757
URMBase Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0756.5
RMBase Model=Qwen-3-1.7B, RL Algorithm=GRPO, Number of generations=2, Epochs=32026.0748.5