Olicorne
Olicorne
I consider myself [pretty experienced with LLMs](https://github.com/thiswillbeyourgithub/wdoc/) and have been using them in my projects for a while. My 2 cents is that only accepting structured output will work better...
Thanks. Then as an **extremely grateful** user of paperless-ngx, I'll manage to find some time in the next few days to take a look at your code in more details...
Hi, I've taken a look at the code as promised. Unfortunately I couldn't do it in one setting so it's been spread out over days and only on a phone...
Also I believe you might want to write some function that makes sure to remove the "thinking" part of the answer if present as i believe there is no universal...
The related issue is #188
> does your prompt not display line breaks in the input box? It looks like line breaks will be replaced with spaces. Yes it does and even if it did...
Well this is markdown *rendering*, which is not necessarily an issue : displaying the # as a header is not a bug and the LLM still sees #. My issue...
Well that's interesting but still a bug: no rendering settings should modify what's sent to the llm, especially not removing information.
You're absolutely right @nhorton , sorry for my inattentiveness. > So basically, Tiptap sees and , doesn't recognize them as part of its schema, and just removes them. It's sanitizing...
Looks like it but a picture would help to be sure.