Metraipolis › archive › dc_dadb670fd9c6

Exploring the Design Space of Permutation Diffusion LLMs — arXiv:2610.20635

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.20635
License (as archived)
CC0-1.0
Archived
2026-10-16T09:06:11.487Z
SHA-256 of the archived text
e7e371bc365401ccc45803b891af02d52307fbebc6ccdfd4344db78d615429f4
Stable URL of this page
https://metraipolis.com/documento/dc_dadb670fd9c6

How to cite

Exploring the Design Space of Permutation Diffusion LLMs — arXiv:2610.20635 — arXiv (en), https://arxiv.org/abs/2610.20635 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dadb670fd9c6 (Metraipolis archive dc_dadb670fd9c6, sha256 e7e371bc365401ccc45803b891af02d52307fbebc6ccdfd4344db78d615429f4)
BibTeX
@misc{dcdadb670fd9c6,
  title = {Exploring the Design Space of Permutation Diffusion LLMs — arXiv:2610.20635},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20635},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dadb670fd9c6; sha256 e7e371bc365401ccc45803b891af02d52307fbebc6ccdfd4344db78d615429f4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20635. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Permutation generation with language models requires both text-conditioned ranking and a global one-to-one assignment between candidates and output positions. We study this problem through three diffusion-LLM parameterizations: hard-assignment prediction over shuffled candidate-index sequences, soft-assignment prediction over doubly stochastic matrices, and direct ordering-score prediction followed by sorting. The three families place permutation constraints at different stages of prediction and decoding and therefore support different training objectives and sampling procedures. We evaluate these designs on recommendation reranking across three Amazon domains with candidate sets of size 10 and 20. At one function evaluation, the hard- and soft-assignment variants achieve Recall@1 comparable to a constrained masked-diffusion baseline, whereas direct ordering-score prediction guarantees valid permutations without Hungarian assignment or Sinkhorn normalization. Additional function evaluations do not consistently improve ranking quality, and their effect depends strongly on the state representation and training objective. These results show that corruption, prediction parameterization, objective, and decoder should be designed jointly for permutation diffusion with language-model denoisers.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.20635 — 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 2026-10-16T08:44:17.852Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).