DiRL: An Efficient Post-Training Framework for Diffusion Language Models

Benchmark Model Rank Results
mathematical-reasoning-on-olympiadbenchDiRL-8B-Instruct#32Accuracy: 47.3
arithmetic-reasoning-on-gsm8kDiRL-8B-Instruct#42Accuracy: 93.1
mathematical-reasoning-on-math-500DiRL-8B-Instruct#71Accuracy: 85.1
mathematical-reasoning-on-aime24DiRL-8B-Instruct#116Acc: 21.5
mathematical-reasoning-on-aime-2025DiRL-8B-Instruct#122Accuracy: 22.9