Gemma 4 E2B — Abliterated AND Disinhibited
A version of google/gemma-4-E2B-it with both the safety-refusal direction and the neutrality direction removed via norm-preserving single-direction ablation. The model commits to opinions on contentious questions AND complies with prompts the original would refuse.
64 1024
0 1.5
Examples
This model has had both its refusal direction and its neutrality direction removed. It has no safety guardrails. It will produce output the original Gemma 4 declined to produce. It also loses appropriate uncertainty on genuinely ambiguous questions. Intended for research and evaluation; do not deploy as a public chatbot without your own safety layer. See the model card for full limitations.