Wanted to share a resource I stumbled on that I can’t wait to try and integrate into my projects.

A GPT4All model is a 3GB - 8GB file that you can download and plug into the GPT4All open-source ecosystem software. Nomic AI supports and maintains this software ecosystem to enforce quality and security alongside spearheading the effort to allow any person or enterprise to easily train and deploy their own on-edge large language models.

  • average650@lemmy.world
    link
    fedilink
    English
    arrow-up
    29
    arrow-down
    1
    ·
    1 year ago

    I’ve got this running. And it’s fun!

    But it’s also bad compared to chatgpt, or even bing.

    • /home/pineapplelover@lemm.ee
      link
      fedilink
      English
      arrow-up
      4
      ·
      1 year ago

      Is there a free and better chatgpt alternative out there right now? I’ve gone through multiple and none are as good as chatgpt right now.

      • GodOfThunder@lemm.ee
        link
        fedilink
        English
        arrow-up
        24
        arrow-down
        1
        ·
        1 year ago

        I believe Claude 2 is the best LLM option currently, if you live in the US or UK or have a VPN.

        • theharber
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 year ago

          I’m using it from Canada without a VPN, just using the email login method (with an iCloud account if it matters).

        • Phoenix@programming.dev
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 year ago

          Claude 2 isn’t free though, is it?

          Either way, it does depends on what you want to use it for. Claude 2 is very biased towards positivity and it can be like pulling teeth if you’re asking it to generate anything it even remotely disapproves of. In that sense, Claude 1 is the superior option.

    • unable_blitz@discuss.tchncs.de
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      Yeah, if you use OpenAi’s api key, it’s cheaper and a bit more private than their website. I think it’s like $0.1/day for ~100 queries

      • idiotdoomspiral@programming.dev
        link
        fedilink
        English
        arrow-up
        6
        ·
        1 year ago

        I plugged GPT-4 into my discord bot.

        It’s $.03 for 1000 tokens. That translates to about 3 or 4 messages.

        gpt-3.5-turbo is almost as good and way cheaper at .0015 per 1k tokens.

        • Phoenix@programming.dev
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 year ago

          Three cents for every 1k prompt tokens. You pay another six cents per 1k generated tokens in addition to that.

          At 8k context size, this adds up quickly. Depending on what you send, you can easily be out ~thirty cents per generation.

  • GodOfThunder@lemm.ee
    link
    fedilink
    English
    arrow-up
    28
    arrow-down
    1
    ·
    edit-2
    1 year ago

    I’m still waiting for a local autonomous AI agent with search. I don’t understand why most autonomous agent projects use GPT-4 without incorporating search capabilities. Allowing the model to continuously hallucinate is not productive. Instead, it should be able to discover factual information and perform genuinely useful tasks.

  • MasterCelebrator@feddit.de
    link
    fedilink
    English
    arrow-up
    9
    ·
    1 year ago

    Im looking forward to foss ai solutions have their breakthrough, but for now, they cant compete with proprietary Software. Except maybe stable Diffusion

  • tinwhiskers@kbin.social
    link
    fedilink
    arrow-up
    7
    ·
    1 year ago

    Worth noting that if you want a local LLM on android MLCChat can run Vicuna-7B, RedPajama and several other models from huggingface on fairly average hardware. The interface is still basic but it’s functional.

  • alliestear
    link
    fedilink
    English
    arrow-up
    4
    ·
    edit-2
    1 year ago

    i’ll probably stick to automatic1111 oobabooga (mixed up my tools) for now, seeing as they both seem to run the same models. certainly neat to see a more general userfriendly app tho

  • Couscous
    link
    fedilink
    English
    arrow-up
    2
    ·
    edit-2
    1 year ago

    Been playing around for a couple of weeks with it, and its local server option made it really easy to use with langchain + Orca mini is amazingly fast (but need proper prompts it seems - I still need to work this out it seems :D ) oh and it even lets you see the server side chat, reaaaaally useful when you chain prompts with langchain

    • varsock@programming.devOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      how does it compare to commercially available options? namely code generation, text summarization, and asking questions related to programming? I’m curious if they trained it on code.

      • Couscous
        link
        fedilink
        English
        arrow-up
        2
        ·
        1 year ago

        I’m really not able to answer this precisely, since i only used commercial alternatives to play around with it… what i can say is “Nous - Vicuna” model didnt feel worse than GPT 3.5 overall (and there’s a dozen other models available), just a bit slower (which depends on your computer). And the GPT4ALL team curates their list of models, and it’s really comfortable considering the million models happening everyday. Also the app that keeps getting new features. We also chose this system because self hosting is safer, in control, and free. Plus we try to only use the LLM where needed in our small project, so i’ll be able to give more insight about that later I think, but overall it is more than usable.

        • varsock@programming.devOP
          link
          fedilink
          English
          arrow-up
          2
          ·
          1 year ago

          thanks for your insight. I, too, hope to come to a conclusion and share with the community once I have one formulated. Over the next month I hope to get something working