• Daxtron2
    link
    fedilink
    arrow-up
    32
    ·
    1 month ago

    You’re running a 405b param model on 24gb of VRAM, no shit it’s not gonna work

      • Daxtron2
        link
        fedilink
        arrow-up
        1
        ·
        1 month ago

        Yeah I’m sure that’s how they got it to run at all lol, luckily they’ve fixed a lot of the issues with earlier versions of model runners, I had blue screen running 7b models back then. One of this size might’ve literally started a fire on my computer

      • Daxtron2
        link
        fedilink
        arrow-up
        1
        ·
        1 month ago

        Yeah that’s most likely what they did to get it to run at all, but expecting it to produce more than a single token on that hardware is laughable