Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-25 · high
A new method splits suffix tokens into three zones by role, then carries forward prior denoising results, no retraining required.
Why now: The paper posted to arXiv this August, targeting a cost every diffusion decoding step still pays.
A new decoding method cuts diffusion language model inference time by up to 72.81x on long sequences, with no extra training. Diffusion models decode several tokens per step, but each step still attends to every unfinished suffix token, a cost that grows with sequence length.
Existing speedups keep attention to a local window of suffix tokens. But they re-initialize every suffix token from scratch each timestep, discarding whatever the model already inferred, per the structured suffix modeling paper.
The new method splits the suffix into three zones: local, middle, and tail. Each zone keeps a different token count based on its role in decoding. Instead of wiping suffix representations at every step, the method carries forward the prior step's decoding results, so denoising information accumulates over time.
The method is orthogonal to parallel decoding and KV caching, the two speed tricks diffusion LMs already use. It stacks with both rather than replacing them. The paper tested the method on three diffusion language models at once.
The paper doesn't say how the speedup scales past the sequence lengths it tested. It also doesn't say whether the gains hold on models larger than the three used here.
Each link below shares sources, entities, or timing with this story.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (caching, carry, token); pushes against this story (versus).
Reported by the same outlet (arxiv.org); overlapping topics (decoding, diffusion, model); pushes against this story (but).
Same source domain / Shared topic
Reported by the same outlet (arxiv.org); overlapping topics (decoding, denoising, diffusion, model).
Shared entity: Structured / Same source domain / Earlier coverage
Both cover Structured; reported by the same outlet (arxiv.org); earlier Structured coverage from 2026-08-24.
Both cover Structured; reported by the same outlet (arxiv.org); earlier Structured coverage from 2026-08-13.
Both cover Structured; reported by the same outlet (arxiv.org); earlier Structured coverage from 2026-02-28.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (count, model); pushes against this story (against).
Reported by the same outlet (arxiv.org); overlapping topics (diffusion, step); pushes against this story (but).