Paper Detail
Daniil Plyusov, Alexey Gorbatovski, Alexey Malakhov, Nikita Balagansky, Boris Shaposhnikov, Daria Korotyshova, Daniil Gavrilov
On-policy distillation (OPD) trains a student on prefixes sampled from its own policy while matching a stronger teacher. This addresses the prefix mismatch of offline distillation, but early student rollouts can still be poor, placing teacher supervision on weak or low-quality prefixes. We propose Trust-Region behavior Blending (TRB), a warmup method that replaces the early rollout policy with the closest-to-teacher behavior policy inside a student-centered KL trust region, while keeping the per-prefix reverse-KL OPD loss unchanged. The KL budget is annealed to zero, so training returns to pure student rollouts after warmup. Across two math-reasoning distillation settings, TRB attains the strongest average among the compared methods.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{plyusov2026trust,
title = {Trust-Region Behavior Blending for On-Policy Distillation},
author = {Daniil Plyusov and Alexey Gorbatovski and Alexey Malakhov and Nikita Balagansky and Boris Shaposhnikov and Daria Korotyshova and Daniil Gavrilov},
year = {2026},
abstract = {On-policy distillation (OPD) trains a student on prefixes sampled from its own policy while matching a stronger teacher. This addresses the prefix mismatch of offline distillation, but early student rollouts can still be poor, placing teacher supervision on weak or low-quality prefixes. We propose Trust-Region behavior Blending (TRB), a warmup method that replaces the early rollout policy with the closest-to-teacher behavior policy inside a student-centered KL trust region, while keeping the per},
url = {https://huggingface.co/papers/2605.31159},
keywords = {on-policy distillation, student policy, teacher policy, prefix mismatch, offline distillation, behavior blending, KL trust region, reverse-KL loss, annealing, huggingface daily},
eprint = {2605.31159},
archiveprefix = {arXiv},
}
{}