Is this even still a thing? It seems to be pretty well dead. Poe-API shat the bed, GPT4FREE got shut down and it’s replacement seems to be pretty much non-functional, Proxies are a weird secret club thing (despite being nearly totally based on scraped corporate keys), etc.

    • Ganbat@lemmyonline.comOP
      link
      fedilink
      English
      arrow-up
      2
      arrow-down
      1
      ·
      1 year ago

      I have an R9 380 that I’m never going to be able to replace. Local isn’t really an option.

      • coyotino [he/him]@beehaw.org
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        2
        ·
        1 year ago

        My experience is with gpt4all (which also runs locally), but I believe the GPU doesn’t matter because you aren’t training the model yourself. You download a trained model and run it locally. The only cap they warn you about is RAM - you’ll want to run at least 16gb of RAM, and even then you might want to stick to a lighter model.

        • Ganbat@lemmyonline.comOP
          link
          fedilink
          English
          arrow-up
          2
          ·
          edit-2
          1 year ago

          No, LLM text generation is generally done on GPU, as that’s rhe only way to get any reasonable speed. That’s why there’s a specifically-made Pyg model for running on CPU. That said, one generation can take anywhere from five to twenty minutes on CPU. It’s moot anyway as I only have 8GB ram.

          • coyotino [he/him]@beehaw.org
            link
            fedilink
            English
            arrow-up
            5
            ·
            1 year ago

            I’m just telling you, it ran fine on my laptop with no discrete GPU 🤷 RAM seemed to be the only limiting factor. But yeah if you’re stuck with 8GB, it would probably be rough. I mean it’s free, so you could always give it a shot? I think it might just use your page file, which would be slow but might still produce results?