Gemma-4-26B-A4B does not require 50+ Gb of vram. It is a MoE model so only 4B of active parameters at a time and not as GPU dependent. I can run it on 16gb of vram and ~20gb of DDR5 regular ram for a 8 bit quant.
wget --mirror --convert-links --adjust-extension --page-requisites --no-parent -e robots=off -P ./website "https://lumon-industries.com/"
wget -P ./website/images "https://lumon-industries.com/images/lumon.png"
wget -P ./website/images "https://lumon-industries.com/images/nope.png"
wget -P ./website/images "https://lumon-industries.com/images/100.png"
wget -P ./website/images "https://lumon-industries.com/images/clipboard.png"
wget -P ./website/images "https://lumon-industries.com/images/mde.gif"
wget -P ./website/shaders "https://lumon-industries.com/shaders/crt.vert.glsl"
wget -P ./website/shaders "https://lumon-industries.com/shaders/crt.frag.glsl"
wget -P ./website "https://lumon-industries.com/favicon.ico"
and then comment out <script>
navigator.serviceWorker.register?.('/service-worker.js').catch(() => {})
</script>
in index.html python3 -m http.server