Metraipolis › archive › dc_49e4642a68cd
From Position Risks to Block Survival: Faster Generation for Diffusion Language Models — arXiv:2609.33390
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.33390
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:01:21.923Z
- SHA-256 of the archived text
324c9653c3330446e2a9943b05a73161948146d1becd82023e48abae19472bc4- Stable URL of this page
https://metraipolis.com/documento/dc_49e4642a68cd
How to cite
From Position Risks to Block Survival: Faster Generation for Diffusion Language Models — arXiv:2609.33390 — arXiv (en), https://arxiv.org/abs/2609.33390 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_49e4642a68cd (Metraipolis archive dc_49e4642a68cd, sha256 324c9653c3330446e2a9943b05a73161948146d1becd82023e48abae19472bc4)
BibTeX
@misc{dc49e4642a68cd,
title = {From Position Risks to Block Survival: Faster Generation for Diffusion Language Models — arXiv:2609.33390},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.33390},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_49e4642a68cd; sha256 324c9653c3330446e2a9943b05a73161948146d1becd82023e48abae19472bc4},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33390. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Diffusion language models (DLMs) can accelerate generation by predicting multiple tokens in parallel, but there is a mismatch between how these tokens are predicted and how they ultimately contribute to generation. Parallel predictions can hardly condition on the tokens selected earlier within the same block, even though their validity depends on this realized prefix. Under the popular proposal-verification decoding, this mismatch makes errors highly asymmetric: an early rejection prevents all subsequent proposals from contributing decoding progress. We introduce BRISK-DLM, a framework that addresses both mismatches by optimizing proposal learning and selection for verified progress. BRISK-DLM trains on self-generated sequences, using risk-reward weighting to dynamically prioritize positions by their impact on verified progress and decoding cost. During inference, a lightweight prefix-conditioned corrector reranks existing candidates using previously selected tokens and preferences distilled from the model's own verifier. The corrector reuses the backbone's parallel representations and requires no additional backbone evaluation, while fused execution keeps its overhead small. BRISK-DLM improves end-to-end throughput by up to 37.4% while preserving task quality, establishing a new quality-throughput frontier for DLM generation. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.33390 — 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 2026-10-07T20:29:02.632Z 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).