RLHF Checkpoints from Clipping Free Policy Optimization for Large Language Models