>Magistral 24b Q8 storywriting tests
It's a weird model. It does a lot of thinking and drafts of the stories. The writing is pretty good usually, but then suddenly some chapters will be very short, basically repeating my prompt just in a few more words and proper punctuation. It can also forget to close its think tag, especially later in context, so whatever it drafted kind of ends up being the final product. And then there's that weird \boxed{\text{whatever}} stuff.

It has potential but doesn't really work properly. Can the thinking be disabled, or does it make it worse? I didn't touch the system prompt yet, it's whatever came standard from the ollama library