<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Profiling on DRM HSE</title><link>https://www.drmhse.com/tags/profiling/</link><description>Recent content in Profiling on DRM HSE</description><generator>Hugo</generator><language>en</language><lastBuildDate>Wed, 05 Aug 2026 09:15:35 +0300</lastBuildDate><atom:link href="https://www.drmhse.com/tags/profiling/index.xml" rel="self" type="application/rss+xml"/><item><title>Tuning CosyVoice3 in Rust: Where the Speed Actually Was</title><link>https://www.drmhse.com/posts/tuning-cosyvoice3-rust-metal/</link><pubDate>Wed, 05 Aug 2026 08:30:00 +0300</pubDate><guid>https://www.drmhse.com/posts/tuning-cosyvoice3-rust-metal/</guid><description>&lt;p>I spent two rounds of work optimising a stage that accounts for 4.4% of runtime, because my own profiler told me it was 17%.&lt;/p>
&lt;p>That is the most useful thing I can tell you about this project, so it goes first. The rest — RTF 2.96 down to 0.697 on the same 132-word passage, a second implementation of a model I had &lt;a href="https://www.drmhse.com/posts/running-funaudio-on-mac-mlx-pytorch/">already made fast once in PyTorch&lt;/a>, this time in Rust on &lt;a href="https://github.com/huggingface/candle"target="_blank" rel="noopener noreferrer" class="external-link">candle&lt;svg class="inline-block w-3 h-3 ml-0.5 align-baseline" fill="none" stroke="currentColor" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-hidden="true">
 &lt;path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M10 6H6a2 2 0 00-2 2v10a2 2 0 002 2h10a2 2 0 002-2v-4M14 4h6m0 0v6m0-6L10 14">&lt;/path>
 &lt;/svg>
&lt;/a> — is downstream of getting the instrument right.&lt;/p></description></item></channel></rss>