Metraipolis › archive › dc_723fd0bcbcf0
Distillation as Probability Transport: Routed On-Policy Distillation — arXiv:2609.08337
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.08337
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T12:57:54.679Z
- SHA-256 of the archived text
5efe9d0da532fd09ce7bc44bbe6d8455ec0b7c235e303f308626d6ab9bd5bda2- Stable URL of this page
https://metraipolis.com/documento/dc_723fd0bcbcf0
How to cite
Distillation as Probability Transport: Routed On-Policy Distillation — arXiv:2609.08337 — arXiv (en), https://arxiv.org/abs/2609.08337 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_723fd0bcbcf0 (Metraipolis archive dc_723fd0bcbcf0, sha256 5efe9d0da532fd09ce7bc44bbe6d8455ec0b7c235e303f308626d6ab9bd5bda2)
BibTeX
@misc{dc723fd0bcbcf0,
title = {Distillation as Probability Transport: Routed On-Policy Distillation — arXiv:2609.08337},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.08337},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_723fd0bcbcf0; sha256 5efe9d0da532fd09ce7bc44bbe6d8455ec0b7c235e303f308626d6ab9bd5bda2},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.08337. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
On-policy distillation (OPD) transfers teacher knowledge on student-generated trajectories, but efficient sampled objectives reduce the teacher distribution to scalar credit on individual tokens. Such credit indicates whether a token should gain or lose probability, yet leaves the corresponding redistribution unspecified. We recast OPD as teacher-guided probability transport and propose RouteOPD (Routed On-Policy Distillation), which decomposes local teacher--student disagreement into student-excess sources and teacher-deficit destinations and couples them into explicit transport pairs. RouteOPD optimizes pairwise log-odds toward jointly realizable targets obtained from a bounded teacher potential, while adapting the transport budget to the concentration of teacher demand. This formulation directs updates toward teacher-preferred destinations and controls their magnitude within a single transport operator. Experiments across four teacher--student settings and four mathematical-reasoning benchmarks demonstrate that RouteOPD consistently outperforms sampled reverse-KL OPD, with improvements accompanied by higher routing fidelity and lower background leakage. These results demonstrate the effectiveness of explicitly modeling probability transport in on-policy distillation. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL). Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2609.08337 — 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 2026-10-16T12:30:25.349Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).