Entity · paper

Pass the Baton: Trajectory-Relayed On-Policy Distillation

paperactiveprovisionalpass-the-baton-trajectory-relayed-on-policy-distillation-dda1f6d0·1 events·first seen 38h ago

Aliases: Pass the Baton: Trajectory-Relayed On-Policy Distillation

Co-occurring entities

on-policy distillation Qwen3.5-0.8B Qwen3-1.7B Qwen3-4B-Instruct FastOPD

More like this (12)

Routing-based On-Policy Distillation on-policy distillation On-Policy Distillation (OPD)On-Policy Co-Distillation Weak-to-Strong Generalization via Direct On-Policy Distillation On-Policy Distillation for LLM Safety: A Routing Approach to Template-Robust Realignment Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation Rethinking Classifier-Free Guidance in On-Policy Diffusion Distillation Multi-Teacher On-Policy Distillation Explainable Reinforcement Learning via Physics-Aware Policy Distillation on-policy self-distillation Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation

Recent events (1)

5arXiv · cs.AI·38h ago·source ↗

Relay-OPD addresses prefix failure in on-policy distillation via teacher-student trajectory handoff

Researchers introduce Relay On-Policy Distillation (Relay-OPD), a training method that addresses 'prefix failure' in on-policy knowledge distillation, where student models compound early reasoning errors throughout a trajectory. The approach detects divergence points where teacher and student continuations asymmetrically diverge, then briefly hands generation to the teacher to produce a corrective 'relay leg' before the student resumes. Evaluated on eight mathematical reasoning benchmarks using Qwen3-4B-Instruct-2507 as teacher and Qwen3-0.6B/1.7B as students, Relay-OPD outperforms standard OPD by +5.73% and the strongest baseline FastOPD by +1.49% on average for the 1.7B model, while also reducing training trajectory length by over 50%.

Open Weights Progress Alignment and RLHF on-policy distillation Qwen3.5-0.8B Pass the Baton: Trajectory-Relayed On-Policy Distillation +3 more