Dominic
Dominic
You can rapidly load and unload ollama models via the api endpoint as suggested above, use keepalive 0 to immediately unload said model.
Btw, check out this pull: https://github.com/Skyvern-AI/skyvern/pull/1160 Seems like the OP has issues with ollama based vision models for some reason
That is sick
What is this
https://status.diekieboy.com/ They have updated their status page to use something that only records incidents from the last 35 mins. No more hetrix
Ollama does provide other vision models, have you tried llava or the 90B version of llama3.2 vision?
> Yes, 90B is just as clueless as 11b sadly. I had the same results. That's unfortunate, could it be a quantization issue? Ollama models usually comes in Q4, perhaps...
Best of luck!
> @rcmorano Qwen2-VL does not work with Ollama or llama.cpp yet afaik. Ollama supports inference of custom models through the ‘ollama create’ endpoint More info on their docs