Metraipolis › archive › dc_bcbabbee58f5

How code helps different tasks? A decompositional lens on LLM post-training — arXiv:2609.33845

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.33845
License (as archived): CC0-1.0
Archived: 2026-10-07T20:57:13.648Z
SHA-256 of the archived text: 5d954a7c43931ab23781689c41e310ca3989ea95a137c2df7dec4376a93f130f
Stable URL of this page: https://metraipolis.com/documento/dc_bcbabbee58f5

How to cite

How code helps different tasks? A decompositional lens on LLM post-training — arXiv:2609.33845 — arXiv (en), https://arxiv.org/abs/2609.33845 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bcbabbee58f5 (Metraipolis archive dc_bcbabbee58f5, sha256 5d954a7c43931ab23781689c41e310ca3989ea95a137c2df7dec4376a93f130f)

BibTeX

@misc{dcbcbabbee58f5,
  title = {How code helps different tasks? A decompositional lens on LLM post-training — arXiv:2609.33845},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.33845},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bcbabbee58f5; sha256 5d954a7c43931ab23781689c41e310ca3989ea95a137c2df7dec4376a93f130f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33845. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Evaluating code data as a single corpus can obscure which types of code data benefit which models and downstream tasks. Effective data selection requires understanding both the benefits of individual categories and whether these benefits persist when categories are combined. We introduce a decompositional lens for studying these effects in LLM post-training. We first decompose an execution-verified code corpus into interpretable categories based on the computational patterns of its solutions. Through controlled fine-tuning experiments, we compare individual categories with a balanced mixture across instruction-tuned models on question answering, mathematics, and code generation. The resulting response maps reveal recurring gains in average question-answering performance, while the same category can improve one model or task and degrade another. The best-performing category also varies with the starting model and target task. We then compose compact mixtures guided by these results and examine whether benefits observed in individual categories persist under joint training. On selected model--task pairs, mixtures whose constituents each improve the target task outperform both their best constituent and full-corpus training while using roughly 10--15\% of the full corpus. These exploratory findings illustrate a \emph{less is more} pattern and highlight how the value of code data in post training depends on which categories are combined for which model and task.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.33845 — 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 2026-10-07T20:29:01.379Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).