this post was submitted on 13 Mar 2024
89 points (88.7% liked)
Asklemmy
43851 readers
706 users here now
A loosely moderated place to ask open-ended questions
Search asklemmy ๐
If your post meets the following criteria, it's welcome here!
- Open-ended question
- Not offensive: at this point, we do not have the bandwidth to moderate overtly political discussions. Assume best intent and be excellent to each other.
- Not regarding using or support for Lemmy: context, see the list of support communities and tools for finding communities below
- Not ad nauseam inducing: please make sure it is a question that would be new to most members
- An actual topic of discussion
Looking for support?
Looking for a community?
- Lemmyverse: community search
- sub.rehab: maps old subreddits to fediverse options, marks official as such
- !lemmy411@lemmy.ca: a community for finding communities
~Icon~ ~by~ ~@Double_A@discuss.tchncs.de~
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I've been using GPT4All on my laptop and using mostly 7B models due to my RAM limitations and I am amazed how good some of them are.
It's been really easy to use. There are models you can download from within the UI or you can get adventurous and download them from elsewhere, they just need to be in the .gguf format. I get most from TheBloke on hugging face.
So far my favourite has been solar-10.7B-instruct-v1.0-uncensored, it has been astonishingly good.
Oooh, do tell me more, please. I've been toying with the idea of setting up gpt4all myself, but I haven't really had the time to look into it very much yet. I have a couple of questions, though:
Yes it runs on Linux, my laptop is running Manjaro and I installed it from the AUR. I'm not sure if the scripting is possible, there is an openAI compliant web API you can turn on so maybe possible through that, you would probably have to feed in the content of the site with the prompt though, I'm not sure there is a better way but I guess that sort of behaviour is a bit out of scope for GPT4All.
There is a local documents feature that allows it to access text files on your machine that you give it specific access to but I think it's fairly limited in its ability.
The GPT services out there use something called 'tools'.
They get presented to the model and the model can 'call' a tool with arguments, which can then extract some data and input it into the context for the model to continue.
I found out, the models which can run on a normal PC (or even a Laptop) are okay, but not super great. (around or a bit worse than ChatGpt3)
The good stuff (e.g. Nous-Capybara 31B or the Mistral/Mixtral ones) needs some more memory and compute.