Bridged Self-Distillation for On-Policy Learning

Benchmark Model Rank Results
mathematical-reasoning-on-hmmt-feb-2025Qwen3-4B + BSD (Nemotron-3-30B-A3B teacher) (extra data)#17Accuracy: 47.5
mathematical-reasoning-on-aime24Qwen3-4B + BSD (Qwen3-32B teacher) (extra data)#37Acc: 79.2
mathematical-reasoning-on-aime-2025Qwen3-4B + BSD (Qwen3-32B teacher) (extra data)#64Accuracy: 68.3