Version 2 using a larger dataset and different training settings.
I think this one gives much better results in some testing on the base Pony model.
Lora for Sekhmet from Re:Zero.