Rethinking Training-Inference Mismatch in LLM Reinforcement Learning: Where It Arises and How to Correct It

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kQwen3-30B-A3B + CIS#16Accuracy: 95.6
mathematical-reasoning-on-olympiadbenchQwen3-30B-A3B + CIS#29Accuracy: 48.62
mathematical-reasoning-on-minerva-mathQwen3-30B-A3B + CIS#32Accuracy: 34.31