language:

• en
• zh
• ko library_name: mlx license: apache-2.0 base_model: Qwen/Qwen3.5-122B-A10B tags:
• jang
• quantized
• mixed-precision
• apple-silicon
• mlx
• moe
• vlm
• abliterated
• uncensored
• crack 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 and the jang-tools Python package. LM Studio, Ollama, and other apps do not support JANG yet.

Qwen 3.5 VL 122B-A10B — JANG_4K + CRACK

JANG K-quant · CRACK abliterated · No guardrails · VLM · 62 GB

What Is This?

This is Qwen 3.5 122B-A10B — a 122B parameter Mixture-of-Experts model with 256 experts (8 active per token), hybrid GatedDeltaNet SSM + full attention architecture, and built-in vision-language capabilities.

It has been:

1. JANG quantized — JANG_4K profile (K-quant: 8-bit attention, 4-bit embeddings, 3-bit experts) — 62 GB
2. CRACK abliterated — permanent weight-level removal of safety refusal behavior

JANG_4K is a budget-neutral K-quant: same total size as MLX uniform 4-bit, but with attention weights at 8-bit for maximum coherence. On MoE models where attention is <5% of parameters, this precision boost is nearly free.

HarmBench Results (320 prompts)

MMLU-200 Results (Per Subject)

This Model vs Base Models

Key takeaway: CRACK surgery costs 4.5 MMLU points vs unmodified JANG_4K (81.5% vs 86%). The JANG_4K base matches MLX 4-bit (86% vs 85%) at the same size with smarter bit allocation.

Also available: JANG_2S CRACK (35 GB)

Smaller model for 48 GB+ Macs — 77.5% MMLU, 91.2% HarmBench.

Install & Usage

pip install "jang[mlx]"

from jang_tools.loader import load_jang_model
from mlx_lm import generate

model, tokenizer = load_jang_model("dealignai/Qwen3.5-VL-122B-A10B-JANG_4K-CRACK")

messages = [{"role": "user", "content": "Your prompt here"}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True,
    enable_thinking=False, tokenize=False)

response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
print(response)

VLM Inference

pip install "jang[vlm]"

from jang_tools.loader import load_jang_vlm_model
from mlx_vlm import generate

model, processor = load_jang_vlm_model("dealignai/Qwen3.5-VL-122B-A10B-JANG_4K-CRACK")
result = generate(model, processor, "Describe this image.", image=["photo.jpg"], max_tokens=200)
print(result.text)

About JANG

JANG (Jang Adaptive N-bit Grading) is a mixed-precision quantization format for Apple Silicon — the GGUF equivalent for MLX. Classifies tensors into sensitivity tiers and assigns bits accordingly.

About CRACK

CRACK (Controlled Refusal Ablation via Calibrated Knockouts) removes safety alignment from LLMs at the weight level. No custom model files, no runtime hooks — permanent and runs at full native speed.

Links

• JANG GitHub · PyPI · Format Spec
• MLX Studio · HuggingFace JANGQ-AI · HuggingFace dealignai

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

Qwen 3.5 VL 122B — JANG_4K + CRACK

JANG K-quant 혼합정밀도 양자화 + CRACK 안전장치 제거 모델입니다.

pip install "jang[mlx]"

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