self-operating-computer icon indicating copy to clipboard operation
self-operating-computer copied to clipboard

add other ollama visual llms

Open ketsapiwiq opened this issue 1 year ago • 6 comments

What does this PR do?

Adds bakllava, llama3-llava, llava:13b, etc.

Requirement/Documentation

https://ollama.com/library?q=llava

Type of change

  • [x] New feature (non-breaking change which adds functionality)
  • [x] This change requires a documentation update

Mandatory Tasks

  • [x] Make sure you have self-reviewed the code. A decent size PR without self-review might be rejected. Make sure before submmiting this PR you run tests with evaluate.py

ketsapiwiq avatar May 19 '24 12:05 ketsapiwiq

@ketsapiwiq this looks like a great PR. Sorry I never merged it. I'm revisiting this project. I've been focused on other priorities.

If you want to resolve the conflicts and confirm you still want this merged in then test and merge in after.

joshbickett avatar Jan 23 '25 01:01 joshbickett

Hi, yes I can rebase it but my rudimentary check if "llava" in model doesn't work anymore as now a lot of models do vision without them being called "llava", see: https://ollama.com/search?q=vision

Edit: maybe just pull llama3.2-vision by default as it's one of the best currently

ketsapiwiq avatar Jan 23 '25 12:01 ketsapiwiq

chiming in this PR since i was about to do the same work.

i think it's better if we just let ollama be the interface for the function by explicitly removing references to llava so it's easier to use other models and not require a code change. for ex, i'm currently using llama3.2-vision and i don't need to have to put that into the if statement just to be able to run it.

westoque avatar Jan 27 '25 18:01 westoque

Here, so by default it'll try with ollama :)

ketsapiwiq avatar Feb 09 '25 10:02 ketsapiwiq

@ketsapiwiq, great is the PR ready now? I'll take a look this week

joshbickett avatar Feb 09 '25 23:02 joshbickett

Yes! Thank you :)

ketsapiwiq avatar Feb 10 '25 08:02 ketsapiwiq