1. 22
    1. 2

      I thought it was going to be using local hardware but lost interest when I saw it requires an API key. How’s the state of the art currently for that use case?

      1. 3

        If you have a multimodal model running locally, say for example with llamafile, ollama or lmstudio, you can easily configure llm to connect to them instead. The llm package is more like a Swiss Army knife to connect which whatever model (local or remote) you like and interact with it

        1. 2

          I’m working on plugins for that at the moment. The local vision models have got pretty good - I’m impressed by Llama 3.2 and Phi 3.5 Vision.