Metraipolis › archive › dc_e8e2fd219c50

OC-OPD: Observation-coordinated Supervision for On-Policy Distillation — arXiv:2611.01965

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.01965
License (as archived)
CC0-1.0
Archived
2026-11-03T10:27:55.260Z
SHA-256 of the archived text
f5d14a67faa3a9d3d23744705d6de765e22bacaad0208e9da994ae1ca7d91dbc
Stable URL of this page
https://metraipolis.com/documento/dc_e8e2fd219c50

How to cite

OC-OPD: Observation-coordinated Supervision for On-Policy Distillation — arXiv:2611.01965 — arXiv (en), https://arxiv.org/abs/2611.01965 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e8e2fd219c50 (Metraipolis archive dc_e8e2fd219c50, sha256 f5d14a67faa3a9d3d23744705d6de765e22bacaad0208e9da994ae1ca7d91dbc)
BibTeX
@misc{dce8e2fd219c50,
  title = {OC-OPD: Observation-coordinated Supervision for On-Policy Distillation — arXiv:2611.01965},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.01965},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e8e2fd219c50; sha256 f5d14a67faa3a9d3d23744705d6de765e22bacaad0208e9da994ae1ca7d91dbc},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.01965. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reverse and forward Kullback--Leibler (KL) divergences provide complementary supervision in on-policy distillation: reverse KL refines the student's active token rankings, and forward KL allocates mass to teacher-supported candidates. At student-generated prefixes, we form a four-observation matrix by treating the student and teacher as both candidate sources and evaluators. The diagonal self-observations measure uncertainty, while source-centered cross-observations recover the two KL objectives. We establish an exact identity equating the Fisher inverse-metric coupling of their logit gradients with the sum of these contrasts, yielding a common local descent direction for distinct positive distributions. Observation-coordinated on-policy distillation (OC-OPD) uses observations to set preferences, closed-form bargaining to coordinate gradients, and distributional discrepancy to scale the update. Across three heterogeneous student settings (non-thinking generation, thinking generation, and base-model initialization), OC-OPD achieves strong gains in mathematical reasoning and transfers effectively to out-of-distribution scientific question answering and code generation. Our code is publicly available at: this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Optimization and Control (math.OC).
Pagina del lavoro: https://arxiv.org/abs/2611.01965 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-03T10:19:50.135Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).