• percent@infosec.pub
    link
    fedilink
    arrow-up
    6
    ·
    3 days ago

    I’ve been using a local Qwen3.8 model as a NixOS sysadmin lately for a non-critical VPS. So far, it has been working surprisingly well! I guess it’s easier for AI agents to gather context when the whole system and apps are defined in code, and immutable when applied.

    • TehWorld@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 days ago

      I’ve been wanting to do this, even as a testbed. I just want computers to do cool shit, and am getting tired of being a Linux sysadmin in my free time. How large of a model do you have assigned to this?

      • percent@infosec.pub
        link
        fedilink
        arrow-up
        2
        ·
        edit-2
        1 day ago

        Just Qwen3.8-27B. That little model is surprisingly smart.

        NixOS also gives you the ability to roll back to previous config versions. I have not needed to do that (yet), but it reduces some risk.

        • TehWorld@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          15 hours ago

          “Just” and a 27B model don’t exactly belong in the same sentence if you ask my wallet. 💸

          • percent@infosec.pub
            link
            fedilink
            arrow-up
            2
            ·
            8 hours ago

            Yeahhh sorry 😕

            I was fortunate to have bought my graphics card almost a year ago, and for a nice discount (a Black Friday sale, IIRC). I’d love to buy another, but they now cost >2x the price I paid.

            FWIW, you can probably run it on CPU if you have enough RAM. I assume it will be really slow though.

            Might also be worth trying Qwen3.6 35B-A3B. It’s not as “smart” as Qwen3.8 27B, but can be much faster on limited resources.