---
格式版本: 2
标题: "Optimize Your Sampling: Tuned Diffusion Sampling with Bayesian Optimization"
原文链接: "https://arxiv.org/abs/2608.18040"
发布日期: "2026-08-18"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Tue, 18 Aug 2026 17:33:00 UTC (7,263 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T15:57:07+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T15:54:04+08:00"
入库时间: "2026-08-20T07:57:07.693Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=inference%20networking&searchtype=all"
匹配关键词:
  []
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 10
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "正文为扩散模型采样优化论文，与超节点/AI Rack/机柜级AI基础设施无关，无任何相关技术或商业信号。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T16:03:14+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 3
AI技术细节: 2
AI商业部署信号: 0
AI完整性: 0
AI摘要: "本文提出OYS方法，将扩散模型采样中时间步的选择视为黑箱优化问题，直接用贝叶斯优化优化目标指标。OYS无需额外训练，在文生图、图像修复等任务上优于默认调度和Align Your Steps；"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:21:56.178Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18040"
---

## Computer Science > Machine Learning

## Title:Optimize Your Sampling: Tuned Diffusion Sampling with Bayesian Optimization

[View PDF](https://arxiv.org/pdf/2608.18040) [HTML (experimental)](https://arxiv.org/html/2608.18040v1)

> Abstract:Sampling from a diffusion model typically requires many forward passes through a large neural network, making generation computationally expensive. While much work has focused on efficient solvers and samplers, comparatively little attention has been paid to selecting the sampling timesteps themselves. A recent line of work optimizes theoretically derived surrogates for sample quality rather than the quality metric itself. We propose Optimizing Your Sampling (OYS), which instead treats timestep selection as a black-box optimization problem, optimizing the target metric directly with Bayesian optimization. OYS outperforms both the default schedules and those of Align Your Steps on text-to-image generation, and improves over the default schedules on inpainting and other image tasks, in both quantitative and human evaluations. OYS requires no additional training, is applicable even to distilled models, and improves both simple and sophisticated samplers such as Euler and DPM-Solver++. A 5-step OYS schedule retains 89%-94% of the quality of a 50-step schedule while reducing inference cost by 10x.

| Subjects: | Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV) |
| --- | --- |
| Cite as: | [arXiv:2608.18040](https://arxiv.org/abs/2608.18040) \[cs.LG\] |
|  | (or [arXiv:2608.18040v1](https://arxiv.org/abs/2608.18040v1) \[cs.LG\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18040](https://doi.org/10.48550/arXiv.2608.18040) |

## Submission history

From: Travis Zhang \[[view email](https://arxiv.org/show-email/6ab0159d/2608.18040)\]  
**\[v1\]** Tue, 18 Aug 2026 17:33:00 UTC (7,263 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18040) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))