Pass the Baton: Trajectory-Relayed On-Policy Distillation

In 30 Sekunden

Was passiert ist

Relay-OPD improves student model training by having a teacher model take over when the student commits to wrong reasoning directions, then resuming student training on corrected trajectories.

Warum es zählt

Matters for engineers training smaller language models on reasoning tasks who want better performance with less compute spent on misdirected generations.

Achtung

Method requires detecting failure points and managing teacher intervention budget; unclear how well trigger detection generalizes across different model sizes and task domains.