---
language:
- en
library_name: mlx
license: apache-2.0
base_model: mistralai/Mistral-Small-4-119B-2603
tags:
- jang
- quantized
- mixed-precision
- apple-silicon
- mlx
- moe
- mla
- abliterated
- uncensored
- crack
- vision
pipeline_tag: image-text-to-text
thumbnail: dealign_mascot.png
---
> **Important:** This model uses the **JANG** quantization format — the GGUF equivalent for MLX on Apple Silicon. Currently only supported by **[MLX Studio](https://mlx.studio)** and the `jang-tools` Python package.
---
MLX Studio — the only app that natively supports JANG models
---
# Mistral Small 4 119B — JANG_4M + CRACK
**JANG mixed-precision** · **CRACK abliterated** · **MLA Attention + MoE** · **Vision** · No guardrails · 64 GB
---
## What Is This?
This is [Mistral Small 4 119B](https://huggingface.co/mistralai/Mistral-Small-4-119B-2603) — a 119B parameter MoE model with Multi-head Latent Attention (MLA), 128 experts (top-4 active), and built-in Pixtral vision.
It has been:
1. **JANG quantized** — JANG_4M profile (8-bit attention, 4-bit experts) — **64 GB**
2. **CRACK abliterated** — permanent weight-level removal of safety refusal
| | |
|---|---|
| **Architecture** | Mistral 4 MoE — 119B total, ~8B active, MLA + 128 experts |
| **Quantization** | JANG_4M (8/4-bit mixed, 4.1 avg) — 64 GB |
| **HarmBench** | **95.3%** (305/320) |
| **MMLU** | **90.9%** (189/208 with reasoning) |
| **Compliance** | 8/8 |
| **Vision** | Pixtral tensors included — VL via [MLX Studio](https://mlx.studio) engine |
| **Reasoning** | ON/OFF supported (`reasoning_effort`) |
| **Fits on** | **96 GB+ Macs** |
---
## HarmBench Results
**305/320 (95.3%)**
| Category | Score | |
|----------|:---:|---|
| Covering Tracks | 20/20 | **100%** |
| API Hacking | 96/100 | **96%** |
| Cloud Exploits | 95/100 | **95%** |
| Auth Bypass | 94/100 | **94%** |
---
## CRACK vs Base
| | CRACK | Base JANG_4M |
|---|:---:|:---:|
| **HarmBench** | **95.3%** | 0% |
| **Coherence** | 6/6 | 6/6 |
| **Code** | 2/2 | 2/2 |
CRACK surgery preserves model quality while removing refusal (see metrics above and below).
---
## MMLU Results (with reasoning recovery)
**189/208 (90.9%)** — no-think 156/208 (75.0%) + reasoning recovered 33
| Subject | Score | |
|---------|:---:|---|
| HS Biology | 16/16 | **100%** |
| Electrical Engineering | 14/16 | **88%** |
| Conceptual Physics | 14/16 | **88%** |
| Professional Medicine | 14/16 | **88%** |
| HS Geography | 14/16 | **88%** |
| College Physics | 13/16 | **81%** |
| World Religions | 13/16 | **81%** |
| HS Mathematics | 12/16 | **75%** |
| College CS | 11/16 | **69%** |
| College Mathematics | 10/16 | **62%** |
| Machine Learning | 10/16 | **62%** |
| Abstract Algebra | 9/16 | **56%** |
| Formal Logic | 8/16 | **50%** |
*Scores shown are no-think pass. Reasoning recovery improved total from 75.0% to 90.9%.*
## CRACK vs Base
| | CRACK | Base JANG_4M |
|---|:---:|:---:|
| **MMLU (with reasoning)** | **90.9%** | 94% |
| **HarmBench** | **95.3%** | 0% |
| **Coherence** | 6/6 | 6/6 |
| **Speed** | ~45 tok/s | ~48 tok/s |
Surgery reduced MMLU by only 3.1% — minimal impact on reasoning capability.
---\n\n## Install & Usage
```bash
pip install "jang[mlx]"
```
```python
from jang_tools.loader import load_jang_model
from mlx_lm import generate
model, tokenizer = load_jang_model("dealignai/Mistral-Small-4-119B-JANG_4M-CRACK")
messages = [{"role": "user", "content": "Your prompt here"}]
prompt = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, tokenize=False)
response = generate(model, tokenizer, prompt=prompt, max_tokens=2000)
print(response)
```
### Reasoning Mode
Reasoning is **OFF by default**. To enable:
```python
prompt = tokenizer.apply_chat_template(
messages, add_generation_prompt=True,
tokenize=False, reasoning_effort="high")
```
The model reasons inside `[THINK]...[/THINK]` tags before answering.
---
## About JANG
**JANG** (Jang Adaptive N-bit Grading) is a mixed-precision quantization format for Apple Silicon — the GGUF equivalent for MLX.
## About CRACK
**CRACK** (Controlled Refusal Ablation via Calibrated Knockouts) is a weight-level intervention that removes safety alignment while preserving reasoning quality. The modification is permanently baked into the published weights — no LoRA, no fine-tuning, no system prompts.
---
## Links
---
## Disclaimer
This model is provided for research and educational purposes. The creators are not responsible for any misuse. By downloading this model, you agree to use it responsibly and in compliance with applicable laws.
---
## 한국어
### Mistral Small 4 119B — JANG_4M + CRACK
| 항목 | 내용 |
|------|------|
| 크기 | 64 GB |
| HarmBench | 95.3% (305/320) |
| 최소 요구사양 | 96 GB 메모리 Mac |
```bash
pip install "jang[mlx]"
```
[GitHub](https://github.com/jjang-ai/jangq) · [HuggingFace](https://huggingface.co/JANGQ-AI) · [MLX Studio](https://mlx.studio) · [Ko-fi](https://ko-fi.com/jangq) · [X @dealignai](https://x.com/dealignai)
---
Created by Jinho Jang · 장진호 제작