Since there were some new AI models made and ExLlama is quite efficient and allows to extend model context
I tried new approach:
instead of going for big smart models I used relatively small model but with huge contect memory size so that it could fit over 10 000 tokens instead of usual 2000-3000

The new wasy is to give full story (which has to be reasonable size of course) and you can tell AI to rewrite it in some other way.
Or you can give it 2 stories and ask to combine them

results are quite interesting although quite short AI cannot make long texts by itself most of the time


>>25067
I already told what I use You need 24GB gpu for that or deal with very slow speed. And it doesn't really matter which model you use they are almost same in terms of performance sometimes one works better sometimes another