English(EN) Denoising Surface: Modeling and Predicting Inference Cost for Diffusion LLM Serving

新的去噪工作负载表面改进了 dLLM 推理成本预测

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-03 04:00

研究人员引入了去噪工作负载表面（DWS），以更准确地对服务期间扩散大语言模型（dLLM）的推理成本进行建模和预测。传统的成本代理对于 dLLM 来说是不够的，因为它们无法捕捉其生成过程的二维结构，该过程涉及具有异构成本的输出块和去噪步骤。DWS 将此结构保留为概率表面，从而能够使用轻量级的、仅提示的预测器来有效估算成本。这种方法在实际服务实验中显示出显著的改进，将成本预测误差降低了高达 2.50 倍，并将端到端延迟降低了高达 1.92 倍。 AI

影响这种新方法可能导致 dLLM 部署的更有效的资源分配和更低的延迟。

排序理由该集群包含一篇学术论文，详细介绍了用于扩散大语言模型推理成本建模和预测的新方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 5 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇学术论文，详细介绍了用于扩散大语言模型推理成本建模和预测的新方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, infra; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Haoyu Zheng, Fangcheng Fu, Binhang Yuan, Yongqiang Zhang, Liang Deng, Hao Wang, Yuanyuan Zhu, Xiao Yan, Jiawei Jiang · 2026-10-03 04:00

Denoising Surface: 扩散式大语言模型推理成本的建模与预测

arXiv:2610.00499v1 Announce Type: new Abstract: As diffusion large language models (dLLMs) become more capable, they are moving from research settings to real-world \textit{serving}, where request management (such as scheduling and resource allocation) relies on accurate estimati…

本文如何被排名

报道来源 [1]

Denoising Surface: 扩散式大语言模型推理成本的建模与预测

相关实体

相关话题