Preprint2026
Pass the Baton: Trajectory-Relayed On-Policy Distillation
Haolei Xu, Xiaowen Xu, Haiwen Hong, et al.
Relay-OPD improves on-policy distillation by detecting student prefix failures and letting the teacher briefly take over to construct relay trajectories, boosting accuracy and cutting training length by over 50%.
0Jul 28, 2026ReasoningBenchmarks
arXiv