Pass the Baton: Trajectory-Relayed On-Policy Distillation

За 30 секунд

Что произошло

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.

Почему это важно

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

На что обратить внимание

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