🇦🇺𝕄𝕦𝕟𝕥𝕖𝕕𝕔𝕣𝕠𝕔𝕕𝕚𝕝𝕖

  • 15 Posts
  • 684 Comments
Joined 9 months ago
cake
Cake day: March 20th, 2024

help-circle








  • Im on a framwork 16 with the cheaper amd cpu, 64gb ram, no dedicated gpu yet tho waiting till that matures a bit (so i can run large llms locally) before i buy one. Performance is great especially with amd since they tend to have many cores. Most things require very little resources and those that dont i can just passthrough more resources. Once uve eaten the virtualization overhead (its surprisingly little) and the multiple instances of linux overhead (also minimal nowdays) the performance is pretty much no different from bare metal u just gotta pass through more resources where u need em (gaming/coding/cad qubes etc).

    Qubes is pretty smart ie i can give a qube basicly no initial memory (as long as its enough to fit all monitors worth of pixels in memory) and then give it a large max memory allocation and it will only use what it needs it wont use the entire allocation if it doesnt need it. Same as disk everything (unless u configure otherwise) is a thin volume so u can allocated more storage to a vm than u even have available to the system as a whole.

    Btw got split password managment working. God i hate systemd.


  • Found some python code that can act as a openai api proxy. Wrote an ai agent with langchain search tools persistent memory image generators sandboxed code execution environment calculator etc. Use an uncensored llm via openrouters, openai api for vector embedding, locality hosted stable diffusion. Locally hosted searxng. Then i can use any openai api compatable frontend and just point it at my proxy and have an advanced agent in almost any environment.

    Havnt published it yet was thinking i could give it an autogpt or baby agi backend but havnt had any success with that yet.




  • A lot of fucking around getting it set up in a way i like. Modifying dom0 scripts live, injecting python themes to get dark mode in the qube manager, fucking around compiling dubious drivers from source and installing them in dom0, fuckig around with getting windows to behave itself (havnt got it working but i just have a drive partition i pass back and forth to tranfer data). Other than that its pretty good. I3 is a godsend for efficiency, have got ctr+shift+c/v as muscle memory now.

    Am currently fucking around with getting split backend for keepass so i can use the keepassxc browser plugin and have it then requesting access to the keepass backend in a seperate qube. Got it working in 1 qube manually. Currently having systemd kicking my ass over a ncat socket in /run/user/1000/app/org.keepassxc… that doesnt exist untill the browser extension tries to access it but thats well after boot so systemd has already crashed the service.

    Overall having lots of fun tinkering. And god do i love the ability to just blow up my os whenever i feel like it. I can sudo rm -rf / --no-preserve-root whenever i feel like it and my system is fine.


  • Yeah ur right. Everything going on here must go through .world as it is the community. Which makes it even weirder that we dont have .world users federaring out.

    My understanding of federation is ur events get sent to ur instance (kbin.life) ur instance sends those events to the instance of the relevent community (lemmy.world) the instance of the community sends that event to all instances subscribed to that community (lemm.ee, kbin.life, etc) my instance receives that event then notified me.

    So this conversion is goibg through .world just that .world users events seem not to be gettibg sent to .ee

    I think we need someone smarter than both of us to fix this