Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding

Benchmark Model Rank Results
mathematical-reasoning-on-math-500R1-Qwen (32B) with CoRD (heterogeneous teachers)#31Accuracy: 94.8
mathematical-reasoning-on-aime24R1-Qwen (32B) with CoRD (heterogeneous teachers)#33Acc: 79.6
mathematical-reasoning-on-aime-2025R1-Qwen (32B) with CoRD (heterogeneous teachers)#56Accuracy: 70.2