<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Nemotron on danilchenko.dev</title><link>https://www.danilchenko.dev/tags/nemotron/</link><description>Recent content in Nemotron on danilchenko.dev</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Fri, 17 Jul 2026 08:25:00 +0000</lastBuildDate><atom:link href="https://www.danilchenko.dev/tags/nemotron/index.xml" rel="self" type="application/rss+xml"/><item><title>Nemotron 3: How a Mamba-Transformer Hybrid Runs 1M Tokens</title><link>https://www.danilchenko.dev/posts/nemotron-3-mamba/</link><pubDate>Fri, 17 Jul 2026 08:25:00 +0000</pubDate><guid>https://www.danilchenko.dev/posts/nemotron-3-mamba/</guid><description>Nemotron 3 swaps most attention layers for Mamba-2 and MoE to reach 1M-token context at 3.3x throughput. Here is how the hybrid architecture actually works.</description></item></channel></rss>