Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kOPDLM-MATH-8B-Thinking#37Accuracy: 93.8
mathematical-reasoning-on-math-500OPDLM-MATH-8B-Thinking#46Accuracy: 92.4
mathematical-reasoning-on-aime24OPDLM-MATH-8B-Thinking#83Acc: 50.0
multi-task-language-understanding-on-mmlu-proOPDLM-8B#107Accuracy: 53.7
mathematical-reasoning-on-aime-2025OPDLM-8B (think@eval)#129Accuracy: 19.4
multiple-choice-question-answering-mcqa-on-gpqa-diamondOPDLM-8B#130Accuracy: 36.1