https://neowutran.ovh/qubes/articles/gaming_linux_hvm.html
Is it outdated? Will this work on the latest Qubes OS version? I want to create an AppVM for LLMs
Note: This version may not be the most up to date. The latest version can usually be found on the Qubes OS forum.
Can i even do GPU passthrough with AppVM to use LLMs or you currently can only do it with HVMs?
You can only passthrough PCI devices to HVM qubes, but app qubes (AppVMs) can be HVM as well.
Will i lose some Qubes OS features(being able to copy paste between qubes) even if its AppVM just because its using HVM?
No.
How dangerous is it to do GPU passthrough this way? If i want to use LLMs to obfuscate my language(rephrase my sentences) to make linguistic analysys impossible, are pros outweighting cons? I can also use some distant VPS to host LLM, but this idea sounds terrible, since it will break stream isolation in Tor and connect all my identites together
I think it’s the same as with other qubes with attached PCI devices (sys-net/sys-usb):
Frequently asked questions (FAQ) | Qubes OS
Device handling security | Qubes OS
The term is adversarial stylometry:
Adversarial stylometry - Wikipedia (on my Wikiless instance)