Introduces POW3R, a policy-aware rubric reward framework that adapts criterion weights during training to improve reinforcement learning performance by empha…
机构:Scale AI
来源:arXiv 2605.20164 | AI4Papers 论文推荐平台