[Feature Request] Add Closed-Loop Critic Refinement and Alignment

Kimi K2 paper proposed _Closed-Loop Critic Refinement and Alignment_ (During RL training, the critic model is refined using verifiable signals).

[https://arxiv.org/html/2507.20534v1#:~:text=Closed-Loop Critic Refinement and Alignment](https://arxiv.org/html/2507.20534v1#:~:text=Closed%2DLoop%20Critic%20Refinement%20and%20Alignment)

The paper argued
"Consequently, this holistic alignment yields comprehensive performance improvements across a wide spectrum of domains, including user intent understanding, creative writing, complex reasoning, and nuanced language comprehension."

Are there any planned add methods like these in the future?