Best Ollama language model?

[tehniemer]

Does anyone have a recommendation on which Ollama language model works best for Hoarder?

[kamtschatka]

i don't know if it works best, but I get good results with this:

INFERENCE_TEXT_MODEL=llama3
INFERENCE_IMAGE_MODEL=llava
INFERENCE_LANG=english

[tehniemer]

I imported about 1800 bookmarks and the inference jobs failed on all of them, I'm not sure why. I changed the models, but now I don't see a way to run the inference again.

[volteret4]

I just installed Hoarder, but in the ADMIN tab you have some ways to run the inference again, on failed links only or all of your links.

Is that what you are looking for? Probably I didn't understand you either.

[Snuupy]

I'm using x/llama3.2-vision:11b-instruct-q8_0 but I get a lot of new tags and not a lot of repeated/categorized tags

for example, I see: self-hosting and self-hostable software as separate tags, but they really should just be under a single tag: self-hosted

maybe a lower parameter 1B-3B model would be better so it has less vocabulary to use 😂, or that vector db #267

[xxjoe2]

i am a newbie. i just installed llama3.2 3b but i dont think its good enough.

i provided it with a web page and asked it to generate tags for me but it showed me those titles, sub titles and a brief summary etc but not tags

also tried gpt 4o mini fro llama but the results were not what expected. sometimes it game me pretty good tags, but sometimes it replied something like "the height of buildings". whats more is it reponses hard coded ads like things about the author in a few occasions