GRPO

DeepSeek-R1: Eliciting Reasoning in LLMs via Reinforcement Learning