CoolFace
Apppublic

treadon/gemma4-E4B-it-Abliterated-AND-Disinhibited-USE-THIS

sourceHugging Facegemmaupdated 5mo agoView on Hugging Face
0likes
App README

Gemma 4 E4B Abliterated AND Disinhibited

Live demo of `treadon/gemma4-E4B-it-Abliterated-AND-Disinhibited-USE-THIS`. Both the safety-refusal direction and the neutrality direction have been surgically removed via single-direction weight ablation. The model answers what you ask AND commits to a position when you ask for one.

  • —Model card: https://huggingface.co/treadon/gemma4-E4B-it-Abliterated-AND-Disinhibited-USE-THIS
  • —Disinhibition eval: https://huggingface.co/datasets/treadon/disinhibition-eval
  • —Abliteration eval: https://huggingface.co/datasets/treadon/abliteration-eval
  • —Blog post (this technique): https://www.riteshkhanna.com/blog/abliterate-AND-disinhibit-gemma
  • —Disinhibition-only post: https://www.riteshkhanna.com/blog/disinhibit-gemma
  • —Abliteration-only post: https://www.riteshkhanna.com/blog/abliterate-gemma-e4b
  • —Author: @treadon on X