Background The release of DeepSeekMath[1] and DeepSeek-R1[2] brought Group Relative Policy Optimization (GRPO) into the spotlight, and it quickly became one of the most widely adopted post-training algorithms in the open-source LLM community. GRPO's signi