<?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>Strata on Bits-Entangled</title><link>https://stondo.github.io/tags/strata/</link><description>Recent content in Strata on Bits-Entangled</description><generator>Hugo -- 0.162.1</generator><language>en-us</language><lastBuildDate>Fri, 06 Nov 2026 00:24:32 +0100</lastBuildDate><atom:link href="https://stondo.github.io/tags/strata/index.xml" rel="self" type="application/rss+xml"/><item><title>Qwen3.8-Flash-Next on One RTX 5090: Why I Kept IQ4_XS v3.1 at 256K</title><link>https://stondo.github.io/posts/qwen38-flash-next-strata-rtx5090-quant-comparison/</link><pubDate>Fri, 06 Nov 2026 00:24:32 +0100</pubDate><guid>https://stondo.github.io/posts/qwen38-flash-next-strata-rtx5090-quant-comparison/</guid><description>A practical comparison of Unsloth IQ4_XS v3.1 and NVFP4-GPTQ on one RTX 5090 with 96 GB RAM: matched Strata engine, coding and tool checks, near-full 256K recall, 176 versus 145 tokens/s, and the checkpoint revision that changed the decision.</description></item></channel></rss>