Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models

Open paper
Benchmark Model Rank Results
mathematical-reasoning-on-aime-2025DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling–Accuracy: 43.2
mathematical-reasoning-on-aime-2026DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling–Accuracy: 50.6
mathematical-reasoning-on-math-500DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling–Accuracy: 95.4

Rank counts only results with a code link.