AI RESEARCH

Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation

arXiv CS.CL

ArXi:2605.11739v1 Announce Type: new On-policy distillation (OPD) has emerged as an efficient post-