Metraipolis › archive › dc_fc94ce4ecbaf

Beyond Prompt Count: How Data Shapes Transfer in On-Policy Distillation — arXiv:2609.37377

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.37377
License (as archived): CC0-1.0
Archived: 2026-10-07T16:23:50.415Z
SHA-256 of the archived text: 9fce7912bdf847e878290ec4041c99f0fd3d3f9537c9ce84386242a8c19215e8
Stable URL of this page: https://metraipolis.com/documento/dc_fc94ce4ecbaf

How to cite

Beyond Prompt Count: How Data Shapes Transfer in On-Policy Distillation — arXiv:2609.37377 — arXiv (en), https://arxiv.org/abs/2609.37377 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fc94ce4ecbaf (Metraipolis archive dc_fc94ce4ecbaf, sha256 9fce7912bdf847e878290ec4041c99f0fd3d3f9537c9ce84386242a8c19215e8)

BibTeX

@misc{dcfc94ce4ecbaf,
  title = {Beyond Prompt Count: How Data Shapes Transfer in On-Policy Distillation — arXiv:2609.37377},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.37377},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fc94ce4ecbaf; sha256 9fce7912bdf847e878290ec4041c99f0fd3d3f9537c9ce84386242a8c19215e8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.37377. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

On-policy distillation (OPD) trains students using teacher feedback on their own sampled responses, yet how prompt choice shapes transfer across teacher-student pairs remains poorly understood. We systematically study prompt quantity, source, and selection across RL- and SFT-continuation pairs and cross-model settings. We find that OPD can be highly prompt-efficient: a few prompts can approach large-pool performance, with four DAPO prompts matching the observed mathematics score of 3,840 DeepMath prompts. However, prompt utility is relational rather than intrinsic: changing only the teacher can reverse the relative effectiveness of mathematics and code prompts. To characterize these transfer differences, we analyze parameter and functional changes across prompt supports and model pairs. Functional alignment with the teacher varies across supports and target tasks; in continuation pairs, teacher-aligned prediction changes can coexist with weak parameter alignment. Continued OPD on effective supports can restore performance after unfavorable transfer. Finally, targeted selection does not consistently outperform uniform random sampling, and filtering out a source that performs poorly alone yields no consistent gain across three paired support draws. Overall, our results distinguish prompt efficiency from prompt interchangeability and show that effective data choice depends on the teacher-student pair and target capability, with random sampling providing a competitive baseline in the studied settings.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.37377 — 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 2026-10-07T15:54:53.623Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).