LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning

Open paper
Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kLYNX (QwQ-32B)#27Accuracy: 94.77
mathematical-reasoning-on-math-500LYNX (QwQ-32B)#54Accuracy: 91.4
mathematical-reasoning-on-aime24LYNX (QwQ-32B)#56Acc: 66.67

Rank counts only results with a code link.