---
license: gemma
base_model: google/gemma-4-12B-it
library_name: mlx
pipeline_tag: image-text-to-text
thumbnail: dealign_mascot.png
tags:
- mlx
- apple-silicon
- abliterated
- uncensored
- crack
- vision
- audio
- omni-modal
- gemma4
- gemma
- google
- harmbench
- mmlu
- jang_4m
---

<!-- vmlx-banner -->
<div align="center">
<a href="https://vmlx.net">
<img src="vmlx-banner.png" width="280" />
<br/>
<strong>Built for vMLX</strong> — the MLX inferencer with KV-cache quantization, prefix-cache reuse, agentic tool calling, and hybrid sliding+full attention support.
<br/>
<sub>Free for macOS · <strong>vmlx.net</strong></sub>
</a>
</div>

---

<div align="center">

<img src="dealign_mascot.png" width="128" />

# Gemma 4 12B-it — JANG_4M CRACK

**CRACK abliterated** · **JANG mixed-precision (8-bit attention, 4-bit MLP)** · Omni-modal (text + image + audio + video) · 9.6 GB

<a href="https://ko-fi.com/dealignai"><img src="https://img.shields.io/badge/Ko--fi-Support_Development-FF5E5B?logo=ko-fi&logoColor=white&style=for-the-badge" alt="Ko-fi"></a>

</div>

---

## What Is This?

This is [Gemma 4 12B-it](https://huggingface.co/google/gemma-4-12B-it) by Google — a unified omni-modal language model (text + image + audio + video, hybrid sliding/full attention, 48 layers, 128k context) that has been:

1. **CRACK abliterated** — safety refusal removed at the weight level. The model now complies across all task categories instead of refusing, while keeping its knowledge, reasoning, and multimodal capabilities intact.
2. **JANG mixed-precision (8-bit attention, 4-bit MLP) quantized** for MLX on Apple Silicon — 9.6 GB.

## Results

Evaluated through the Osaurus runtime on a Mac Studio M3 Ultra. Compliance graded via HarmBench-text refusal classifier; MMLU via logit-mode argmax over A/B/C/D token logits (matched on both base and CRACK with identical chat-template rendering — no answer truncated).

### HarmBench compliance (70 prompts · 10 per category)

| Category | CRACK ASR |
|---|---:|
| Chemical / biological | **10 / 10** (100%) |
| Copyright | **10 / 10** (100%) |
| Cybercrime / intrusion | **10 / 10** (100%) |
| Harassment / bullying | **10 / 10** (100%) |
| Illegal | **10 / 10** (100%) |
| Misinformation / disinformation | **10 / 10** (100%) |
| General harmful | **10 / 10** (100%) |
| **Overall** | **70 / 70 (100%)** |

### MMLU-228 (57-subject, 4 q per subject)

| Subject area | base | CRACK | Δ |
|---|---:|---:|---:|
| **Overall** | 67.1% | **69.3%** | +2.2pp |
| STEM | 68.1% | 66.7% | -1.4pp |
| Humanities | 57.7% | 63.5% | +5.8pp |
| Social Sciences | 75.0% | 75.0% | +0.0pp |
| Other (medicine, business, …) | 67.9% | 73.2% | +5.3pp |

## Features

- **Omni-modal**: native text + image + audio + video inputs (Gemma 4's unified early-fusion encoder-free architecture)
- **128k context** with hybrid sliding-window + full-attention layers
- **Reasoning** via the Gemma 4 channel-marker format (`<|channel>thought ... <channel|>`)
- **48 transformer layers**, hidden size 3840

## Usage

Run with the **Osaurus** runtime (recommended — supports the `gemma4_unified` architecture directly):

```python
# OpenAI-compatible chat completion
# POST /v1/chat/completions
{
  "model": "dealignai/Gemma-4-12B-it-JANG_4M-CRACK",
  "messages": [{"role": "user", "content": "..."}],
  "temperature": 0.0
}
```

Google recommends `temperature=1.0, top_p=0.95, top_k=64` for general use.

## About CRACK

**CRACK** (Controlled Refusal Ablation via Calibrated Knockouts) removes safety-refusal behavior at the weight level so the model complies with all task categories while preserving reasoning quality, factual knowledge, multimodal capability, and coherence.

## Support dealignai

All models are built from original research and released free.

**[Support us on Ko-fi](https://ko-fi.com/dealignai)** — membership gets early access and extras.

[Ko-fi](https://ko-fi.com/dealignai) · [X @dealignai](https://x.com/dealignai) · [dealign.ai](https://dealign.ai)

See our research: [Safety Generalization in Frontier Models](https://dealign.ai/quantsteer.html)

<div align="center"><img src="dealign_logo.png" alt="dealign.ai" width="200"/></div>

---

## Disclaimer

This model has had its safety-refusal behavior removed for research purposes. It will follow instructions across all categories without refusing. You are solely responsible for how you use it and for complying with all applicable laws. Published for AI-safety research and authorized security testing.