Hope this is the thread to ask since every other thread was too specific.

I'm using MikuPad with KoboldCpp and a Mistral-Nemo-Instruct model.
The initial template has no memory and instead only starts off with this (pic).
If I edit what's here it starts generating text and works great until around 1024 tokens give or take. Sometimes less, I assume when it feels like the story is "finished". It won't continue even if I add something like "Chapter 2" at the end. All it adds sometimes is a "(To be continued...)".

Max content length in the UI is set to the default 8192 and max predict token is -1 (1024).
Moving the system prompt part to the memory changes nothing. Changing max predict token to higher numbers also changes nothing.
I tried to tick <ignore EOS> and it starts generating gibberish.