מודל שפה כבודק: למידת חיזוק עם הערכת ערך ממצבים פנימיים

תקציר מקורי באנגליתarXiv:2605.07579v3 Announce Type: replace-cross Abstract: Reinforcement learning with verifiable rewards (RLVR) for Large Reasoning Models rests on variance reduction, which requires both a reliable baseline and high prompt diversity within each training batch. This is especially difficult in multi-domain training for general reasoning models, where prompts from different tasks induce highly diverse gradient signals. Existing approaches fall short in different ways: GRPO estimates its baseline as the group mean over rollouts from the same prompt, so an accurate baseline leaves fewer distinct prompts in the batch, while PPO avoids this trade-off by training a policy scale critic, roughly doubling the cost of training. We introduce POISE (Policy Optimization with Internal State Value Estimat