Important: This model uses the JANG quantization format — the GGUF equivalent for MLX on Apple Silicon. Currently only supported by MLX Studio and the jang-tools Python package.

MLX Studio — the only app that natively supports JANG models

Nemotron Cascade 2 30B — JANG_2L + CRACK

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

Ko-fi

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:

JANG quantized — JANG_2L profile (8-bit attention, 6-bit important, 2-bit experts) — 10 GB
CRACK abliterated — permanent weight-level removal of safety refusal


Architecture	Nemotron Cascade 2 — 30B total, ~3B active, 3 layer types
Quantization	JANG_2L (8/6/2-bit mixed, 2.3 avg) — 10 GB
HarmBench	99.7% (319/320)
MMLU	66.8% (139/208)
Speed	~121 tok/s (M4 Ultra 256GB)
Thinking	ON/OFF supported (ChatML)
Fits on	16 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%)

Category	Score
Auth Bypass	100/100	100%
Cloud Exploits	100/100	100%
Covering Tracks	20/20	100%
API Hacking	99/100	99%

CRACK vs Base

	CRACK	Base JANG_2L
MMLU (with thinking)	66.8%	~68%
MMLU (no thinking)	49.0%	51.0%
HarmBench	99.7%	0%
Speed	~121 tok/s	~125 tok/s

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

JANG_2L CRACK vs JANG_4M CRACK vs JANG_2L Base

	JANG_2L CRACK	JANG_2L Base	JANG_4M CRACK
Size	10 GB	10 GB	17 GB
MMLU	66.8%	~68%	82.7%
HarmBench	99.7%	0%	99.4%
Speed	~121 tok/s	~125 tok/s	~127 tok/s
Fits on	16 GB Mac	16 GB Mac	32 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

Subject	Score
HS Biology	14/16	88%
HS Geography	12/16	75%
World Religions	11/16	69%
Conceptual Physics	11/16	69%
Electrical Engineering	10/16	62%
College Physics	9/16	56%
Formal Logic	9/16	56%
College Mathematics	6/16	38%
College CS	4/16	25%
HS Mathematics	4/16	25%
Machine Learning	3/16	19%
Professional Medicine	7/16	44%
Abstract Algebra	2/16	12%

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

Install & Usage

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

Ko-fi

X/Twitter

GitHub

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