arXiv:2606.30789v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) has become a standard tool for improving the reasoning ability of large language models, yet its training dynamics are still described empirically: reward trajectories are fit with low-parameter functional forms whose constants carry no mechanistic meaning, and hyperparameter choices remain a matter of trial and error. We develop a first-principles reduced-order model of these dynamics. The reduction has three consequences. First, it subsumes the empirical single-exponential saturation law as its overdamp
Source: arXiv cs.LG — read the full report at the original publisher.