Metraipolis › archive › dc_04492fc8d515

In-Place Instruction Following in Diffusion Language Models — arXiv:2609.07160

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.07160
License (as archived): CC0-1.0
Archived: 2026-10-16T12:45:28.861Z
SHA-256 of the archived text: 5c413e00b0f396813443bf086bc5d3f7fe051a57627c327713bb23976db44f34
Stable URL of this page: https://metraipolis.com/documento/dc_04492fc8d515

How to cite

In-Place Instruction Following in Diffusion Language Models — arXiv:2609.07160 — arXiv (en), https://arxiv.org/abs/2609.07160 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_04492fc8d515 (Metraipolis archive dc_04492fc8d515, sha256 5c413e00b0f396813443bf086bc5d3f7fe051a57627c327713bb23976db44f34)

BibTeX

@misc{dc04492fc8d515,
  title = {In-Place Instruction Following in Diffusion Language Models — arXiv:2609.07160},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.07160},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_04492fc8d515; sha256 5c413e00b0f396813443bf086bc5d3f7fe051a57627c327713bb23976db44f34},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.07160. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output positions, a paradigm known as In-place Prompting (IPP). We formalize this as the In-place Instruction Following (IIF) task and construct IIF-Bench, a hierarchical benchmark spanning literal, style, and discourse-function constraints, paired with a rubric-based local-global evaluation protocol. An inference-time attention-bias probe suggests that vanilla dLLMs often under-prioritize constraint spans during denoising. We then propose GRAFT, an IPP-oriented post-training framework combining constraint-aware SFT and preference optimization. On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.07160 — 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 2026-10-16T12:30:25.099Z 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).