Coupled Variational Reinforcement Learning for Language Model General Reasoning

Benchmark Model Rank Results
mathematical-reasoning-on-minerva-mathQwen3-14B with CoVRL#27Accuracy: 38.1
multi-task-language-understanding-on-mmlu-proQwen3-14B with CoVRL#88Accuracy: 63.1
mathematical-reasoning-on-math-500Qwen3-14B with CoVRL#105Accuracy: 76.1
mathematical-reasoning-on-aime24Qwen3-14B with CoVRL#131Acc: 9.5