Adaptive Pullback: Reweighting Supervision at Penalty Jumps in On-Policy Distillation
On-policy distillation (OPD) supervises student-generated responses using sampled-token student-teacher log-ratios. A positive log-ratio penalizes a token that the student assigns higher probability to than the teacher, but its current value does not distinguish sustained high penalties from abrupt increases along a response. We examine these local changes by comparing consecutive blocks of tokens within the same response. Our diagnostic identifies penalty jumps, abrupt increases in average penalty that persist on held-out tokens. Across these jumps, penalties rise while teacher confidence decreases. We call this pattern penalty-confidence tension. Boundaries into high-penalty blocks do n...
| Authors | Guangyu Liu, Siyuan Gan, Shilong Zhou et al. |
|---|---|
| Primary category | cs.LG |
| Published | 20 October 2026 |
| Last updated at source | 20 October 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 21 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.