<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Numerical Stability | Mingkai Zheng's Homepage</title><link>https://zhengmk.github.io/tag/numerical-stability/</link><atom:link href="https://zhengmk.github.io/tag/numerical-stability/index.xml" rel="self" type="application/rss+xml"/><description>Numerical Stability</description><generator>Wowchemy (https://wowchemy.com)</generator><language>en-us</language><lastBuildDate>Mon, 28 Sep 2026 00:00:00 +0000</lastBuildDate><image><url>https://zhengmk.github.io/media/icon_hu0158c3d4eaa2a6b730ca31a67c18838d_142122_512x512_fill_lanczos_center_2.png</url><title>Numerical Stability</title><link>https://zhengmk.github.io/tag/numerical-stability/</link></image><item><title>Broken Symmetry in BF16 Attention: Why FlashAttention Gradients Blow Up Late in Training</title><link>https://zhengmk.github.io/publication/broken-symmetry-in-bf16-attention-why-flashattention-gradients-blow-up-late-in-training/</link><pubDate>Mon, 28 Sep 2026 00:00:00 +0000</pubDate><guid>https://zhengmk.github.io/publication/broken-symmetry-in-bf16-attention-why-flashattention-gradients-blow-up-late-in-training/</guid><description/></item></channel></rss>