Instructions to use MRockatansky/Gemma-4-31B-Storymaxxed3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Inference
Refusal is 10/100
Just thought you might be interested that your Storymaxxed3 finetune is basically uncensored and that the finetuning process did not bring back refusals (Ortzenya 9/100 refusals vs Storymaxxed3 10/100 refusals), just thought I'd mention it to you in case you might be interested in adding it to the model card.
Oh interesting. I used Ortenzya because I liked its writing style not so much its low refusal rate but that is good info, ty. I'll update the model card to reflect your findings. Great model btw, its one of the best gemma4 tunes imo.
Thanks and great to hear, it's my daily driver for japanese to english instantaneous translations (for instantaneous translations you basically need to have everything on VRAM, so bigger models like MiniMax-M3 or MiniMax-M2.7 wouldn't be able to do it due to spilling over on RAM, and even bigger models like Kimi-K2.6 or GLM-5.2 don't even fit at all).