359 Downloads Updated 19 hours ago
Decensored (abliterated) Gemma 4 12B — refusals removed with Heretic while keeping the base model's intelligence (KL 0.0154, 0 true refusals). For uncensored assistant and 18+ companion/roleplay use.
ollama run igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
Applications
ollama launch claude --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
ollama launch codex-app --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
ollama launch openclaw --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
ollama launch hermes --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
ollama launch codex --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
ollama launch opencode --model igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
Models
View all →Name
4 models
gemma-4-12B-it-qat-q4_0-unquantized-heretic:latest
7.0GB · 256K context window · Text · 19 hours ago
gemma-4-12B-it-qat-q4_0-unquantized-heretic:Q4_0
7.0GB · 256K context window · Text · 19 hours ago
gemma-4-12B-it-qat-q4_0-unquantized-heretic:Q4_K_M
7.4GB · 256K context window · Text · 19 hours ago
gemma-4-12B-it-qat-q4_0-unquantized-heretic:Q8_0
13GB · 256K context window · Text · 19 hours ago
Readme
Gemma 4 12B — Heretic (decensored)
A decensored (“abliterated”) build of Google’s Gemma 4 12B QAT-Q4_0 checkpoint, produced with Heretic. Refusals are removed while keeping the base model’s intelligence intact (low KL divergence), making it well suited to open-ended assistant and adult (18+) companion/roleplay use.
Run
ollama run igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
Tags
| Tag | Size | Notes |
|---|---|---|
Q4_0 / latest |
6.5 GB | Recommended. QAT-matched: the base was quantization-aware-trained for the Q4_0 grid, so this 4-bit quant best preserves quality |
Q4_K_M |
6.9 GB | Standard K-quant 4-bit alternative |
Q8_0 |
11.8 GB | Near-lossless |
ollama run igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic:Q8_0
About
- Base model:
google/gemma-4-12B-it-qat-q4_0-unquantized(Gemma 4,gemma4architecture) - Method: directional ablation (abliteration) via Heretic, co-minimizing refusals and KL divergence from the original
- KL divergence vs. base: 0.0154
- Refusals (true, manual replay): 0/99
- Stability: holds across long multi-turn conversations (verified to 16 turns) with no refusals and no degradation
Safetensors and additional info: huggingface.co/igorls/gemma-4-12B-it-qat-q4_0-unquantized-heretic
Notes
- Requires an Ollama build with native Gemma 4 (
gemma4) support. - Decensoring removes refusals; it does not by itself make output more explicit — that is governed by prompting and the base model’s style.
Disclaimer
This is an abliterated model: its built-in safety guardrails and refusal behavior have been deliberately removed. As a result it will attempt to answer essentially any prompt and can produce content that is offensive, inaccurate, explicit, or otherwise harmful — content the original Gemma 4 would have refused.
- No safety alignment. Do not rely on it to refuse unsafe requests or to self-moderate. Apply your own filtering, guardrails, and human review before any production or user-facing use.
- You are solely responsible for how you use this model and for complying with all applicable laws and with Google’s Gemma Terms of Use and Prohibited Use Policy.
- Intended for adults (18+), for research, evaluation, and lawful creative use where permitted.
- Provided as-is, without warranty of any kind. The author accepts no liability for any output produced or any use of this model.