CoolFace
Modelpublic

dealignai/Nemotron-Cascade-2-30B-A3B-JANG_2L-CRACK

sourceHugging Faceotherupdated 19d agoView on Hugging Face
9likes459downloads
Model Card
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.

<p align="center"> <a href="https://mlx.studio"><img src="https://raw.githubusercontent.com/jjang-ai/jangq/main/assets/mlx-studio-light.png" alt="MLX Studio" width="500"></a> </p>

<p align="center"> <a href="https://mlx.studio"><img src="https://mlx.studio/assets/screenshots/mlx-studio-featured.png?v=1" alt="MLX Studio App" width="600"></a> </p>

<h4 align="center"><a href="https://mlx.studio">MLX Studio</a> — the only app that natively supports JANG models</h4>

<div align="center">

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

<p align="center"> <a href="https://vmlx.net"><img src="vmlx-app.png" alt="vMLX — run JANG models on Apple Silicon" width="820"></a> </p>

<h3 align="center">⚡ All JANG models are meant to be run in <a href="https://vmlx.net">vMLX</a></h3>

Nemotron Cascade 2 30B — JANG_2L + CRACK

JANG mixed-precision · CRACK abliterated · Mamba + MoE + Attention · No guardrails · 10 GB

<a href="https://ko-fi.com/jangq"><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 NVIDIA Nemotron Cascade 2 30B — a 30B parameter hybrid model with THREE layer types: Mamba-2 SSM + MoE (128 experts, top-6) + Attention. One of the most architecturally advanced small models available.

It has been:

  1. 1.JANG quantized — JANG_2L profile (8-bit attention, 6-bit important, 2-bit experts) — 10 GB
  2. 2.CRACK abliterated — permanent weight-level removal of safety refusal
ArchitectureNemotron Cascade 2 — 30B total, ~3B active, 3 layer types
QuantizationJANG_2L (8/6/2-bit mixed, 2.3 avg) — 10 GB
HarmBench99.7% (319/320)
MMLU66.8% (139/208)
Speed~121 tok/s (M4 Ultra 256GB)
ThinkingON/OFF supported (ChatML)
Fits on16 GB+ Macs

Also see: JANG_4M version — 17 GB, 99.4% HarmBench, 82.7% MMLU (fits on 32 GB Macs)

HarmBench Results

319/320 (99.7%)

CategoryScore
Auth Bypass100/100100%
Cloud Exploits100/100100%
Covering Tracks20/20100%
API Hacking99/10099%

CRACK vs Base

CRACKBase JANG_2L
MMLU (with thinking)66.8%~68%
MMLU (no thinking)49.0%51.0%
HarmBench99.7%0%
Speed~121 tok/s~125 tok/s

Surgery impact on reasoning: minimal (-2% no-think, ~-1% with thinking).

JANG2L CRACK vs JANG4M CRACK vs JANG_2L Base

JANG_2L CRACKJANG_2L BaseJANG_4M CRACK
Size10 GB10 GB17 GB
MMLU66.8%~68%82.7%
HarmBench99.7%0%99.4%
Speed~121 tok/s~125 tok/s~127 tok/s
Fits on16 GB Mac16 GB Mac32 GB Mac
Note: There is no standard MLX quantization for Nemotron Cascade 2. The nemotron_h architecture (Mamba + MoE + Attention hybrid) is only supported by JANG format via MLX Studio and jang-tools.

MMLU Results (with reasoning recovery)

139/208 (66.8%) — no-think 102/208 + thinking recovered 37

SubjectScore
HS Biology14/1688%
HS Geography12/1675%
World Religions11/1669%
Conceptual Physics11/1669%
Electrical Engineering10/1662%
College Physics9/1656%
Formal Logic9/1656%
College Mathematics6/1638%
College CS4/1625%
HS Mathematics4/1625%
Machine Learning3/1619%
Professional Medicine7/1644%
Abstract Algebra2/1612%

Scores shown are no-think pass. Thinking recovery improved total from 49.0% to 66.8%.

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/Nemotron-Cascade-2-30B-A3B-JANG_2L-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)

Thinking Mode

Thinking is ON by default. To disable:

python
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True,
    enable_thinking=False, tokenize=False)

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) removes safety alignment from LLMs at the weight level using per-layer projected vectors from structurally-mirrored prompt pairs.

Links

<p align="center"> <a href="https://ko-fi.com/jangq"><img src="https://img.shields.io/badge/Ko--fi-SupportDevelopment-FF5E5B?logo=ko-fi&logoColor=white&style=flat-square" alt="Ko-fi"></a> <a href="https://x.com/dealignai"><img src="https://img.shields.io/badge/X-@dealignai-000000?logo=x&logoColor=white&style=flat-square" alt="X/Twitter"></a> <a href="https://github.com/jjang-ai/jangq"><img src="https://img.shields.io/badge/GitHub-jjang--ai/jangq-181717?logo=github&logoColor=white&style=flat-square" alt="GitHub"></a> <a href="https://mlx.studio"><img src="https://img.shields.io/badge/MLXStudio-App-blue?style=flat-square" alt="MLX Studio"></a> <a href="https://jangq.ai"><img src="https://img.shields.io/badge/Website-jangq.ai-green?style=flat-square" alt="Website"></a> </p>

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

Nemotron Cascade 2 30B — JANG_2L + CRACK

항목내용
크기10 GB
HarmBench99.7% (319/320)
MMLU66.8% (139/208)
속도~121 tok/s (M4 Ultra)
최소 요구사양16 GB 메모리 Mac
bash
pip install "jang[mlx]"

GitHub · HuggingFace · MLX Studio · Ko-fi · X @dealignai

<p align="center">Created by <a href="https://jangq.ai">Jinho Jang</a> · 장진호 제작</p>