Interesting result you got there. Good job! Finetuning is probably the way to go for consistent results for a given style.

I have been able to get closer to what I want via prompting with one model but not perfectly. For example, I asked a model to write a detailed description of an object in the style of Charles Dickens and although it deviated every so often from the style, overall it felt very “Dickens-esque”. Result quality varied depending on the author I prompted for. But most of the models I have tried (large and small) totally ignored the prompt and wrote bland encyclopedia entries. At the very least, I’d like a model that knows something about classic authors and their works - but I know it’s a big ask.

BTW, if you ever decide to write that book of yours, I’d be interested in reading it!

Yes, please open source this. It is an amazing idea. Thanks for doing this.

One big thing that you (or someone else) could do to make this accessible (and thus more popular) would be to create a “one-click installer”. This would allow those with little to no coding experience to benefit from this (and that’s a lot of people). Or refine the code in such a way that it could work in the background (or could easily be made to work) with any of the existing GUIs currently out there (e.g. LMStudio, OogaBooga, etc.). No idea how easy or hard this would be (as I am only now learning Python) but thought I’d throw it out there. Thanks again for working on this.