Rubric-based On-policy Distillation

Benchmark Model Rank Results
mathematical-reasoning-on-aime24ROPD (Qwen3-4B)#42Acc: 75.41
mathematical-reasoning-on-aime-2025ROPD (Qwen3-4B)#60Accuracy: 68.75
multiple-choice-question-answering-mcqa-on-gpqa-diamondROPD (Qwen3-4B)#105Accuracy: 55.05