• ExtraJudgement@lemmy.world
      link
      fedilink
      English
      arrow-up
      17
      ·
      2 months ago

      I’m really hoping Soofi models are going to be good. Because afaict they’ll be the first (and so far only) open source (not just open weight) models in the ~30B parameters range.

        • story@lemmy.zip
          link
          fedilink
          English
          arrow-up
          5
          ·
          2 months ago

          I’ve been using the 7b version for about 2 weeks now to monitor my basic k3s metrics and send me emails about it!

          Fun Fact, it is difficult to convince this thing it isn’t Claude without a system prompt

          • story@lemmy.zip
            link
            fedilink
            English
            arrow-up
            6
            ·
            2 months ago

            (if you’re about to send me a rude prompt about how this could have been a cron job, i completely agree with you but i already had the dang thing up and i wanted to see if it could do it)

            i have a web ui for it if anyone wants to try

            • Dentzy@sh.itjust.works
              link
              fedilink
              English
              arrow-up
              1
              ·
              2 months ago

              Just because you used “prompt” instead of “post”, for a second I thought that this was the model’s response to you after you asked it for a solution to monitor 😂

        • ExtraJudgement@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          ·
          2 months ago

          Thank you very much, I hadn’t heard about this model coming out anywhere!

          Seems like I’ll need to wait just a bit more for it to be available in Ollama. As soon as it’s there I’ll try it out :)

        • sakphul@discuss.tchncs.de
          link
          fedilink
          English
          arrow-up
          2
          ·
          2 months ago

          Thanks for sharing! Never heard of them but their smaller models (especially the 36B and 32B Model) sound very promising for lcoal deployments.

          However it is kinda strange that the 32B model performs that much worse in Terminal-Bench 2.1 compared to the 36B parameter model.

        • ExtraJudgement@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          ·
          2 months ago

          The good Apertus model is too large at 70B, unfortunately. The 8B model is much too small to be useful for what I need.