When every rollout for a problem fails, GRPO receives neither a successful exemplar nor a nonzero group-relative advantage, so it learns nothing from exactly…
机构:UCSD
来源:arXiv 2610.21018 | AI4Papers 论文推荐平台