Text Generation
Safetensors
English
gemma4
conversational
instruct
finetune
roleplay
creative-writing
style-tune
Instructions to use Gryphe/Gemma-4-31B-StyleTune with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Inference
For next fine-tune
#8
by Hypersniper - opened
On the next fine-tune, try to reduce the number of em "—" dashes the model puts out. I'm not sure what's going on, but it seems like every model overuses them, and it’s a telltale sign that the text was written by an LLM.
Wouldn't banning em dashes in the user UI, i.e. SillyTavern, technically resolve this? Or does it introduce other issues long-term?
mfw regex exists
mfw regex exists
sure if you want grammatical errors