arXiv cs.AI / cs.LG / cs.CL·1d agoPEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning#agentic-rl#prefill-decode#gpu-schedulingAI research