1
Escaping the KL Agreement Trap in On-Policy Distillation
只要输入论文链接或标题，就能生成关联文献图谱，帮研究者快速发现领域脉络与核心工作。
arXiv:2606.09471v1 Announce Type: new Abstract: On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score stude…