Which Ollama LLM's work for you?

[danshaw]

Trying to save a few bucks and have most of the OCR and tagging be done locally. OCR is working for the most part for me on a Mac Studio (32GB RAM) running the minicpm-v model. I am having issues with Ollama stalling at "Stopping..." for over 25 minutes at a time, but that is for another post.

Curious what others are using for their local LLM when it comes to tagging of title, tags, correspondent, etc. I've tried a few, and all of them are very verbose and not able to handle providing me just the title or just the tag. I've even tried to be very explicit in the templates, but alas, OpenAI (GPT4-o) is much more accurate. If you have a pretty solid workflow and don't mind sharing your template, that would be much appreciated.