mlsanityData to 7 Oct 2026

Denoising Surface: Modeling and Predicting Inference Cost for Diffusion LLM Serving

arXiv 2610.00499 ↗First posted 30 Sep 2026cs.LGcited by 0

On the map: Efficient LLM inference and serving → Diffusion LLM and MoE inference

Abstract (arXiv)

As diffusion large language models (dLLMs) become more capable, they are moving from research settings to real-world textitserving, where request management (such as scheduling and resource allocation) relies on accurate estimation of per-request inference cost. However, common cost proxies fall short for dLLMs: output length ignores that one forward pass can unmask multiple tokens, and denoising-step count ignores the textitheterogeneous per-step costs. We observe that the block-autoregressive generation mechanism induces a two-dimensional execution structure over output blocks and within-block denoising steps, whereas these proxies collapse it into a scalar, discarding information essential for characterizing the cost. Motivated by this insight, we propose the Denoising Workload Surface (DWS), which preserves this two-dimensional block-step structure as a probability surface to weight the heterogeneous per-step costs. We then design a coarse-to-fine training scheme that enables a lightweight prompt-only predictor to accurately predict the complex DWS. This predictor runs efficiently even on a single CPU core, avoiding GPU contention with the serving model. Since DWS decouples request-dependent execution behavior from deployment-specific cost factors, the predictor transfers across hardware configurations without retraining. In textitreal-world serving experiments, DWS reduces cost-prediction error by up to 2.50× over scalar-based predictors, while the DWS-guided shortest-job-first scheduler reduces end-to-end latency by up to 1.92× for online chatbots.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Diffusion LLM and MoE inference, one of 12 directions in Efficient LLM inference and serving. Builds on 28 of its 35 references that are in the map. First posted on arXiv 30 Sep 2026.

Builds on

28 of 35 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding15 Dec
  2. 2025Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing8 Aug
  3. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding28 May
  4. 2025dKV-Cache: The Cache for Diffusion Language Models21 May

Other references · 24

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,665
  4. 2020MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformerscited by 1,414
  5. 2025Large Language Diffusion Modelscited by 872
  6. 2024Simple and Effective Masked Diffusion Language Modelscited by 800
  7. 2025Dream 7B: Diffusion Large Language Modelscited by 515
  8. 2023LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Datasetcited by 449
  9. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 385
  10. 2024Scaling Diffusion Language Models via Adaptation from Autoregressive Modelscited by 268
  11. 2025Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inferencecited by 173
  12. 2025LLaDA2.0: Scaling Up Diffusion Language Models to 100Bcited by 154
12 more
  1. 2023Response Length Perception and Sequence Scheduling: An LLM-Empowered LLM Inference Pipelinecited by 86
  2. 2024Efficient LLM Scheduling by Learning to Rankcited by 83
  3. 2023S3: Increasing GPU Utilization during Generative Inference for Higher Throughputcited by 81
  4. 2023Fairness in Serving Large Language Modelscited by 77
  5. 2024Efficient Interactive LLM Serving with Proxy Model-based Sequence Length Predictioncited by 64
  6. 2024Don't Stop Me Now: Embedding Based Scheduling for LLMscited by 34
  7. 2025dInfer: An Efficient Inference Framework for Diffusion Language Modelscited by 33
  8. 2026Scheduling LLM Inference with Uncertainty-Aware Output Length Predictionscited by 7
  9. 2026Robust Length Prediction: A Perspective from Heavy-Tailed Prompt-Conditioned Distributionscited by 3
  10. 2025Ranking Before Serving: Low-Latency LLM Serving via Pairwise Learning-to-Rankcited by 2
  11. 2026TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offloadcited by 2
  12. 2026Optimus: Elastic Decoding for Efficient Diffusion LLM Servingcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models
  2. 2026CAI-DLLM: Convergence Aware Inference for Diffusion Language Models
  3. 2025FlashDLM: Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusion
  4. 2026DiLaServe: High SLO Attainment Serving for Diffusion Language Models
  5. 2026Accelerating Masked Diffusion Large Language Models: A Survey of Efficient Inference Techniques
  6. 2026Diffusion Language Models: An Experimental Analysis
  7. 2025Fewer Denoising Steps or Cheaper Per-Step Inference: Towards Compute-Optimal Diffusion Model Deployment
  8. 2026Leveraging Inference-Time Compute for Diffusion Models via Global Scheduling of Denoising Trajectories