mlsanityData to 7 Oct 2026

DScale: Scaling Block-Diffusion Speculative Decoding with Adaptive Verification

arXiv 2609.37532 ↗First posted 29 Sep 2026cs.DC cs.AIcited by 0

On the map: Efficient LLM inference and serving → Speculative decoding

Abstract (arXiv)

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

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Speculative decoding, one of 12 directions in Efficient LLM inference and serving. Builds on 28 of its 38 references that are in the map. First posted on arXiv 29 Sep 2026.

Builds on

28 of 38 references.

Closest ancestors · 13

Same direction, newest first

  1. 2026DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation6 Jul
  2. 2026Domino: Decoupling Causal Modeling from Autoregressive Drafting in Speculative Decoding28 May
  3. 2026DFlash: Block Diffusion for Flash Speculative Decoding5 Feb
  4. 2025EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test3 Mar
  5. 2024AdaEDL: Early Draft Stopping for Speculative Decoding of Large Language Models via an Entropy-based Lower Bound on Token Acceptance Probability24 Oct
  6. 2024EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees24 Jun
  7. 2024SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths30 May
  8. 2024Break the Sequential Dependency of LLM Inference Using Lookahead Decoding3 Feb
  9. 2024EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty26 Jan
  10. 2024Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads19 Jan
  11. 2023SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verification16 May
  12. 2023Accelerating Large Language Model Decoding with Speculative Sampling2 Feb
1 more
  1. 2022Fast Inference from Transformers via Speculative Decoding30 Nov

Other references · 15

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  3. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  4. 2023Let's Verify Step by Stepcited by 3,598
  5. 2021Program Synthesis with Large Language Modelscited by 3,509
  6. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,107
  7. 2024DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Servingcited by 595
  8. 2024Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Servecited by 507
  9. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 385
  10. 2024MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attentioncited by 375
  11. 2024LayerSkip: Enabling Early Exit Inference and Self-Speculative Decodingcited by 302
  12. 2025FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inferencecited by 121
3 more
  1. 2024DeepSpeed-FastGen: High-throughput Text Generation for LLMs via MII and DeepSpeed-Inferencecited by 106
  2. 2025DOPD: A Dynamic PD-Disaggregation Architecture for Maximizing Goodput in LLM Inference Servingcited by 5
  3. 2025BrownoutServe: SLO-Aware Inference Serving under Bursty Workloads for MoE-based LLMscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Block Verification Accelerates Speculative Decoding
  2. 2026Bastion: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting
  3. 2026LongSpark: Efficient speculative decoding with a fixed-cost parallel drafter
  4. 2026D^2SD: Accelerating Speculative Decoding with Dual Diffusion Draft Models
  5. 2026JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting
  6. 2026SpecFold: Folding Multi-Branch Redundancy for Faster Speculative Decoding in Diffusion Language Models
  7. 2026DFlare: Scaling Up Draft Capacity for Block Diffusion Speculative Decoding
  8. 2025Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMs