Question for a consideration of mine.
I'm curious what the dataset is and if that's not possible, I totally understand. I use your model in a great quantity of my merges and it excels at making longer responses and prose highly improved. Works fantastic with instruct merges etc. Thought I'm aiming to make a new merge type and it's going more NSFW focused.
Is there quality material for that sort of thing in this LoRA? Thank you in advance and also big thank you for making these amazing models. If you release a lovecraft 30b I'd love to make a 13/30b model series release with existential horror as the focus as well.
Appreciate it.
It was just a small selection of books (16MB, very small) from the books3 dataset cleaned up and formatted properly. It's copyrighted stuff so I don't think I can post the dataset. There's actually little to no NSFW in it.
Can i ask how you made the book dataset?
How are the json files formated?
Theres tons of q&A dataset examples and a few guides for them, but none on making book based datasets.
I understand you can not share the dataset itself but just an example of how its formated could be helpful