SPEC-RL: Accelerating On-Policy Reinforcement Learning via Speculative Rollouts Paper • 2509.23232 • Published Sep 27 • 3