• sobchak@programming.dev
    link
    fedilink
    English
    arrow-up
    2
    ·
    9 hours ago

    They’re OK. It’s kinda amazing how much lossy data can be compressed into a 12GB model. They don’t really start to get comparable to the large frontier models until 70B+ parameters, and you would need serious hardware to run those.

    • CheeseNoodle@lemmy.world
      link
      fedilink
      English
      arrow-up
      3
      ·
      5 hours ago

      Oh I mean with setup, like I can download ollama and a basic model fine enough and get a generic chat bot. But if I want something that can scan through PDFs with direct citation (like the Nvidia LLM) or play a character then suddenly its all git repositiories and cringey youtube tutorials.

      • sobchak@programming.dev
        link
        fedilink
        English
        arrow-up
        2
        ·
        4 hours ago

        That would be nice. I don’t know how to do that other than programming it yourself with something like langgraph.