EasyPPO is pitched as more stable PPO for LLM post-training

The author presents EasyPPO as a way to make PPO more stable during LLM post-training by fixing the critic rather than changing the actor update. They report better results and zero training collapse across all their experiments.