525
this post was submitted on 27 Oct 2023
525 points (94.9% liked)
Technology
58816 readers
4669 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
For large context models the hardware is prohibitively expensive.
I can run 4bit quantised llama 70B on a pair of 3090s. Or rent gpu server time. It's expensive but not prohibitive.
How many tokens can you run it for?
3k?Can't recall exactly, and I'm getting hardwarestability issues.
I’m trying to get to the point where I can locally run a (slow) LLM that I’ve fed my huge ebook collection too and can ask where to find info on $subject, getting title/page info back. The pdfs that are searchable aren’t too bad but finding a way to ocr the older TIFF scan pdfs and getting it to “see” graphs/images are areas I’m stuck on.
I personally use runpod. It doesn't cost much even for the high end level stuff. Tbh the openai API is easier though and gives mostly better results.
I specifically said "large context" how many tokens can you get through before it goes insanely slow?
Max token windows are 4k for llama 2 tho there's some fine tunes that push the context up further. Speed is limited by your budget mostly, you can stack GPUs and there are most models available (including the really expensive ones)
I'm just letting you know, If you want something easy, just use ChatGtp. I don't find them overly expensive for what it is.