Datasets:

Huegli

/

Sillitavern_L3_Settings

like 1

It gives interesting effects. I have the impression that larger models (number of parameters) respond even better to these instructions. Unfortunately my pc doesn't allow for more detailed tests, I checked llama 3 70b q2 - and it was painfully slow, but all 3 stories generated looked better than on the standard settings from silly tavern. It's actually very interesting how this affects the model, in stable-diffusion it seems more random - I thought it was similar here, so I didn't worry too much about it... my mistake.