Metraipolis › archive › dc_5bef4ffb4125

DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification — arXiv:2609.37532

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.37532
License (as archived)
CC0-1.0
Archived
2026-10-07T16:27:22.209Z
SHA-256 of the archived text
73a50b4154cf76791b80cfe61fadade3b67042b9aa5fe9715843d2a1e1c2b103
Stable URL of this page
https://metraipolis.com/documento/dc_5bef4ffb4125

How to cite

DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification — arXiv:2609.37532 — arXiv (en), https://arxiv.org/abs/2609.37532 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5bef4ffb4125 (Metraipolis archive dc_5bef4ffb4125, sha256 73a50b4154cf76791b80cfe61fadade3b67042b9aa5fe9715843d2a1e1c2b103)
BibTeX
@misc{dc5bef4ffb4125,
  title = {DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification — arXiv:2609.37532},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.37532},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5bef4ffb4125; sha256 73a50b4154cf76791b80cfe61fadade3b67042b9aa5fe9715843d2a1e1c2b103},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.37532. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Growing large language model applications demand efficient inference. At high concurrency, block-diffusion speculative decoding suffers from verification padding, rejected candidates, and incompatibility between variable prefixes and fixed-shape graphs. Uniform truncation sacrifices acceptable tokens. We present DScale, preserving drafter architecture, weights, and full draft length. A separate 112K-parameter predictor requires neither confidence calibration nor hardware speed-curve preparation. Path-aware tiles reduce padding. Dynamic verify-length (DVL) allocation packs scored prefixes into half the native verification capacity. Fixed-address workspaces propagate changing boundaries through verification and acceptance while reusing captured graphs. On A100-40GB with tensor parallelism 1, Qwen3-8B and Qwen3-4B cover four datasets and concurrency 8-32, reusing each target's frozen predictor. Geometric-mean throughput gains across these configurations are respectively 43.9% and 48.8% over DFlash, 22.2% and 37.7% over DSpark, and 24.4% and 32.0% over Domino, with lower request latency. Cumulative ablations show that adding the three mechanisms successively increases geometric-mean throughput, while budget adjustment improves accepted-token retention. GPU profiling shows that complete decode-step time on GSM8K decreases by 30.8-52.5% relative to DFlash

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.37532 — 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 2026-10-07T15:54:53.991Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).