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

arXiv: https://arxiv.org/abs/2610.00499 · HTML: https://mlsanity.com/paper/2610.00499/

## Key facts

- No citing papers in the map yet (data through 14 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.

## Placed in

- Efficient LLM inference and serving / [Diffusion LLM and MoE inference](https://mlsanity.com/inference/diffusion-llm-and-moe-inference/)

## Builds on (most cited first, 28 of 28)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Structured Denoising Diffusion Models in Discrete State-Spaces](https://mlsanity.com/paper/2107.03006/) (2021)
- [MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers](https://mlsanity.com/paper/2002.10957/) (2020)
- [Large Language Diffusion Models](https://mlsanity.com/paper/2502.09992/) (2025)
- [Simple and Effective Masked Diffusion Language Models](https://mlsanity.com/paper/2406.07524/) (2024)
- [Dream 7B: Diffusion Large Language Models](https://mlsanity.com/paper/2508.15487/) (2025)
- [LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset](https://mlsanity.com/paper/2309.11998/) (2023)
- [Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding](https://mlsanity.com/paper/2505.22618/) (2025)
- [Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models](https://mlsanity.com/paper/2503.09573/) (2025)
- [Scaling Diffusion Language Models via Adaptation from Autoregressive Models](https://mlsanity.com/paper/2410.17891/) (2024)
- [Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference](https://mlsanity.com/paper/2508.02193/) (2025)
- [LLaDA2.0: Scaling Up Diffusion Language Models to 100B](https://mlsanity.com/paper/2512.15745/) (2025)
- [dKV-Cache: The Cache for Diffusion Language Models](https://mlsanity.com/paper/2505.15781/) (2025)
- [Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing](https://mlsanity.com/paper/2508.09192/) (2025)
- [Response Length Perception and Sequence Scheduling: An LLM-Empowered LLM Inference Pipeline](https://mlsanity.com/paper/2305.13144/) (2023)
- [Efficient LLM Scheduling by Learning to Rank](https://mlsanity.com/paper/2408.15792/) (2024)
- [S3: Increasing GPU Utilization during Generative Inference for Higher Throughput](https://mlsanity.com/paper/2306.06000/) (2023)
- [Fairness in Serving Large Language Models](https://mlsanity.com/paper/2401.00588/) (2023)
- [Efficient Interactive LLM Serving with Proxy Model-based Sequence Length Prediction](https://mlsanity.com/paper/2404.08509/) (2024)
- [dInfer: An Efficient Inference Framework for Diffusion Language Models](https://mlsanity.com/paper/2510.08666/) (2025)
- [Don't Stop Me Now: Embedding Based Scheduling for LLMs](https://mlsanity.com/paper/2410.01035/) (2024)
- [ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding](https://mlsanity.com/paper/2512.13586/) (2025)
- [Scheduling LLM Inference with Uncertainty-Aware Output Length Predictions](https://mlsanity.com/paper/2604.00499/) (2026)
- [Robust Length Prediction: A Perspective from Heavy-Tailed Prompt-Conditioned Distributions](https://mlsanity.com/paper/2604.07931/) (2026)
- [Ranking Before Serving: Low-Latency LLM Serving via Pairwise Learning-to-Rank](https://mlsanity.com/paper/2510.03243/) (2025)
- [TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offload](https://mlsanity.com/paper/2605.20179/) (2026)
- [Optimus: Elastic Decoding for Efficient Diffusion LLM Serving](https://mlsanity.com/paper/2605.24832/) (2026)

## 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.