Guide: Self-hosting open source GPT chat with no GPU using GPT4All
Guide: Self-hosting open source GPT chat with no GPU using GPT4All
GPT4All is easy for anyone to install and use. It allows you to download from a selection of ggml GPT models curated by GPT4All and provides a native GUI chat interface. Hardware CPU: Any cpu will work but the more cores and mhz per core the better. RAM: Varies by model requiring up to 16GB I...
Ever time i see a post like this i ask the same thing and i have yet to receive answer.
Why should i care?
There are so many open source language models, all with different strengths and weaknesses. There are tools to run them on any OS with all kinds of different hardware requirements.
This has been the case since before chatgpt came out and has exponentially blown up since.
Gpt4all is just a single recent model. But in recent weeks it always gets the headlight under “run chatgpt at home”
What does it do to stand out? Why would i use this and not one of the vicuna or llama models?
Hugging face has a leaderboard for open source large language models.
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
If you are interested in running this tech at home, familiarize yourself with multiple models because they all will behave differently depending on your hardware and your needs.
2Its a collection of models you can download. It acts as a simple gui entry point into the llm world. Great to test different stuff.
2Dumb question: Is GPT4All a specific model? Is it just the UI for other models? I thought it uses LLaMA under-the-hood, but I'm not sure.
1
Anybody know of a good guide for hosting this with GPU? Every guide seems to be talking about running it on CPU when one would expect the opposite to be true. I have not been able to use my RTX 3060 for this so far.
2Take my answer with a grain of salt, but I'm pretty sure if you have a GPU you can just run the same models and it should work more efficiently for you. The only difference for you is you can run some of the larger models.
1
I tried that. GPT4all is a hog. You'll need at least 16GB of RAM.
0I loved this however my only disappointment is that you can't use it as a server others can connect to and use the chat interface
0Use this webui (its the stabld diffusion ui for llm)
https://github.com/oobabooga/text-generation-webui
I am pretty sure it has a sever option.
Here is a list of the models it likely supports, including gpt4all. https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
Best one i tried is wizard vicuna 13B running on a rtx2070
1oh hey this is super useful, thanks! :D
0
It's a shame it only seems to be at the level of davinci-003 by now. I'm super interested in this, but that's just not good enough for most of the things I use GPT-3/4 for today...
-1Nice!
-1