<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Éthiqueia — Research</title><description>Technical writing across the Éthiqueia research threads: memory systems, cognitive architectures, agent runtimes, commodity training, fractal cognition.</description><link>https://yandesbiens.com/</link><language>en</language><item><title>proof drop #1 — running a 24 GB model on a 24 GB GPU (and the honest catch)</title><link>https://yandesbiens.com/blog/ufm-benchmark/</link><guid isPermaLink="true">https://yandesbiens.com/blog/ufm-benchmark/</guid><description>UFM lets a single RTX 4090 run a routed model whose expert bank doesn&apos;t fit in VRAM. Here&apos;s the benchmark, the code, and the regime where it doesn&apos;t help.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate><category>memory</category><category>ufm</category><category>benchmark</category><category>local-first</category><category>memory</category><category>moe</category><category>rtx4090</category></item></channel></rss>