<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Systems | Ruize Xia · Artificial Minds, Human Values</title><link>https://portfolio.xiaruize.org/tags/systems/</link><atom:link href="https://portfolio.xiaruize.org/tags/systems/index.xml" rel="self" type="application/rss+xml"/><description>Systems</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Sat, 21 Mar 2026 00:00:00 +0000</lastBuildDate><image><url>https://portfolio.xiaruize.org/media/icon_hu_982c5d63a71b2961.png</url><title>Systems</title><link>https://portfolio.xiaruize.org/tags/systems/</link></image><item><title>On-Device Diffusion Kernels</title><link>https://portfolio.xiaruize.org/projects/on-device-diffusion/</link><pubDate>Sat, 21 Mar 2026 00:00:00 +0000</pubDate><guid>https://portfolio.xiaruize.org/projects/on-device-diffusion/</guid><description>&lt;p>This project turns Modulated Diffusion (MoDiff) from an operation-count story into a hardware result. The manuscript reports fused low-bit kernels and a cache-update fusion strategy that cuts extra memory traffic during iterative denoising.&lt;/p>
&lt;p>On the evaluated setup, the implementation reaches up to &lt;strong>1.8×&lt;/strong> runtime speedup over FP32 and up to &lt;strong>42.2%&lt;/strong> lower memory I/O. The public tree lives at
, forked from the official ICML 2025 MoDiff codebase and used as the systems implementation path for the paper.&lt;/p>
&lt;ul>
&lt;li>Paper:
&lt;/li>
&lt;li>Code:
&lt;/li>
&lt;/ul></description></item></channel></rss>