<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>x1zz — blog</title><description>Notes on compilers, Rust, and ML tooling by Sewoo Jang.</description><link>https://x1zz.com/</link><language>en</language><item><title>Three open Burn PRs, and a CI failure that wasn&apos;t mine</title><link>https://x1zz.com/blog/burn-prs-red-main/</link><guid isPermaLink="true">https://x1zz.com/blog/burn-prs-red-main/</guid><description>I had three PRs to tracel-ai/burn sitting unmerged and assumed the problem was my code. It wasn&apos;t — the required Windows check was red on main itself, because a dependency bump re-resolved two versions of the windows crate.</description><pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate><category>rust</category><category>open-source</category><category>burn</category><category>ci</category></item><item><title>Four backends, one missing op: Min/Max scatter in Burn</title><link>https://x1zz.com/blog/burn-pr-5582/</link><guid isPermaLink="true">https://x1zz.com/blog/burn-pr-5582/</guid><description>Element-wise scatter and select_assign only supported Assign, Add, and Mul — Min and Max panicked on every backend. Closing the gap meant matching four architectures and getting the autodiff gradient right.</description><pubDate>Mon, 14 Sep 2026 00:00:00 GMT</pubDate><category>rust</category><category>open-source</category><category>burn</category><category>autodiff</category></item><item><title>My first Burn PR got rejected. Here&apos;s what the maintainers taught me.</title><link>https://x1zz.com/blog/burn-pr-5555/</link><guid isPermaLink="true">https://x1zz.com/blog/burn-pr-5555/</guid><description>PR #5542 was closed because I put a check in the wrong layer. PR #5555 got merged because I finally understood the architecture. The difference was one question.</description><pubDate>Wed, 09 Sep 2026 00:00:00 GMT</pubDate><category>rust</category><category>open-source</category><category>burn</category><category>deep-learning</category></item></channel></rss>