Metraipolis › archive › dc_37786f30417c

OPSRD: On-Policy Self-Role Distillation — arXiv:2609.39884

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.39884
License (as archived)
CC0-1.0
Archived
2026-10-07T12:45:51.954Z
SHA-256 of the archived text
d4f23aba3cbc55409016d4275780a7fb06d5381831ce1ddbc5e069f43ea359f1
Stable URL of this page
https://metraipolis.com/documento/dc_37786f30417c

How to cite

OPSRD: On-Policy Self-Role Distillation — arXiv:2609.39884 — arXiv (en), https://arxiv.org/abs/2609.39884 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_37786f30417c (Metraipolis archive dc_37786f30417c, sha256 d4f23aba3cbc55409016d4275780a7fb06d5381831ce1ddbc5e069f43ea359f1)
BibTeX
@misc{dc37786f30417c,
  title = {OPSRD: On-Policy Self-Role Distillation — arXiv:2609.39884},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.39884},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_37786f30417c; sha256 d4f23aba3cbc55409016d4275780a7fb06d5381831ce1ddbc5e069f43ea359f1},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39884. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Role prompting elicits specialized behavior from large language models through an expert identity, offering a lightweight way to guide reasoning on demanding tasks. However, evaluating or distilling complete role-prompted answers can miss useful next-token preferences when the sampled solution remains incorrect. Transferring these preferences also requires an objective that reaches alternatives the student rarely predicts. We introduce OPSRD, which uses a fixed expert role as privileged teaching context for on-policy self-distillation without reference solutions. A role-free student generates a trajectory, and a frozen instance of the same base model supplies role-conditioned distributions on its exact prefixes, exposing alternatives beyond the sampled continuation. Teacher-weighted forward KL targets alternatives the student underestimates, with clipping to limit individual vocabulary contributions. Supervision is restricted to the highest-entropy half of student positions, concentrating learning where predictions are uncertain. Experiments on three competition-math benchmarks with Qwen3-1.7B, 4B, and 8B show improvements over the base models without role prompts at inference. Forward KL achieves the highest macro-averaged accuracy among the three evaluated divergences at every scale. Code is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.39884 — 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 2026-10-07T12:14:14.166Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).