Uncertainty-Aware Speculative Decoding for Diffusion Language Models in Long Document Generation
Computer Life · 20.07.2026 · 14 (3) · 7-12
DOI: 10.54097/w9c56t64 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer Life, online erschienen am 20.07.2026. Verfasst von Yuna Kim. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Masked diffusion language models (MDLMs) are attractive for long-form generation because they denoise many positions under bidirectional context rather than committing tokens strictly from left to right. The same property makes inference expensive: every reverse step may require dense attention over a long partially masked sequence. Recent sliding-window and speculative wrappers demonstrate that pretrained MDLMs can be accelerated without changing their weights, especially the SW-SpeedDLM framework, which combines segmented sliding-window denoising, cross-segment summary compression, and window-level speculative acceptance. This paper proposes Uncertainty-Aware Speculative Decoding for Diffusion Language Models (UASD-DLM), a complementary inference scheduler for long document generation. Instead of using a fixed speculation depth for every window, UASD-DLM estimates local predictive uncertainty from entropy, margin, and summary-conflict signals, then chooses both the draft depth and context refresh policy dynamically. Low-risk windows receive longer drafts, while high-risk windows are verified earlier to avoid wasted drafting and boundary errors. The acceptance step retains the same rejection-sampling correction as standard speculative decoding, so changing the draft length does not alter the verifier distribution. To avoid unsupported large-model claims, we provide a reproducible CPU proxy benchmark using real text extracted from the uploaded SW-SpeedDLM article and a seeded long-range dependency corpus. In this controlled setting, UASD-DLM improves the speed-memory trade-off over fixed-depth speculative decoding and recovers technical-term accuracy lost by summary-free windowing. The complete code, raw results, and Word manuscript accompany this submission.
Zitieren
APA
Kim, Y. (2026). Uncertainty-Aware Speculative Decoding for Diffusion Language Models in Long Document Generation. Computer Life, 14(3), 7-12. https://doi.org/10.54097/w9c56t64
Vancouver
Kim Y. Uncertainty-Aware Speculative Decoding for Diffusion Language Models in Long Document Generation. Computer Life. 2026;14(3):7-12. doi:10.54097/w9c56t64
BibTeX
@article{Kim2026,
author = {Kim, Yuna},
title = {Uncertainty-Aware Speculative Decoding for Diffusion Language Models in Long Document Generation},
journal = {Computer Life},
year = {2026},
volume = {14},
number = {3},
pages = {7--12},
doi = {10.54097/w9c56t64}
}
EUVIMED