Metraipolis › archive › dc_0b390dfd7356

Parser Transitions Already Reveal Structural Responsibility: KV Persistence for Structured Generation — arXiv:2608.28276

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.28276
License (as archived): CC0-1.0
Archived: 2026-10-17T11:06:37.680Z
SHA-256 of the archived text: 4e7462ae1d5eb90f4131779fbcea347258159f8859c9166b2469f93484a23149
Stable URL of this page: https://metraipolis.com/documento/dc_0b390dfd7356

How to cite

Parser Transitions Already Reveal Structural Responsibility: KV Persistence for Structured Generation — arXiv:2608.28276 — arXiv (en), https://arxiv.org/abs/2608.28276 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0b390dfd7356 (Metraipolis archive dc_0b390dfd7356, sha256 4e7462ae1d5eb90f4131779fbcea347258159f8859c9166b2469f93484a23149)

BibTeX

@misc{dc0b390dfd7356,
  title = {Parser Transitions Already Reveal Structural Responsibility: KV Persistence for Structured Generation — arXiv:2608.28276},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.28276},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0b390dfd7356; sha256 4e7462ae1d5eb90f4131779fbcea347258159f8859c9166b2469f93484a23149},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.28276. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

LLM agents use structured generation for JSON, SQL, and function calls. Constrained decoding ensures valid formats, but KV compression can discard history needed for correct function and argument selection. Structural responsibility can diverge from model saliency: importance scores can overlook generated tokens' structural roles. Parser transitions already computed for legality encode these roles, including required arguments and structural boundaries, providing a free retention signal. We introduce PASK (Parser-Aware Structural KV Persistence), which reuses this signal for KV persistence by layer group without online importance estimation. Task-error sensitivity sets protection floors, while attention-output distortion ranks upgrades under the remaining budget. Offline calibration produces a frozen table for structural lookup and budget accounting. PASK outperforms budget-controlled baselines on function calling and text-to-SQL, improving BFCL accuracy by up to 8.67 pp over the strongest comparator. Against Full KV, it cuts peak GPU memory by 47.5 percent and achieves up to 3.3 times TPOT speedup.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-31 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.28276 — 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 2026-10-17T10:50:25.813Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-31?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).