A key challenge in on-policy distillation (OPD) is extracting effective supervision from teacher knowledge. Supervision from token probabilities may underuti…
机构:浙大
来源:arXiv 2610.23306 | AI4Papers 论文推荐平台