arXiv Artificial Intelligence

DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification

1 hour ago 30.09.2026 07:00 👁 1 views
DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification

Quick summary

arXiv:2609.37532v1 Announce Type: cross Abstract: 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-l

Key takeaways

  • arXiv:2609.37532v1 Announce Type: cross Abstract: 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.

Why it matters

“DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification” should be evaluated beyond branding and benchmark scores. Its practical importance will emerge in task accuracy, latency, unit cost, safety and integration with real workflows.

Kaynak sitede devamını oku: arXiv Artificial Intelligence ↗