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.

Qwen 3.5 VL 35B — JANG_4K + CRACK

JANG mixed-precision · CRACK abliterated · Vision-Language · No guardrails · 18 GB

What Is This?

This is Qwen 3.5 VL 35B — a 35B parameter hybrid SSM/Attention Mixture-of-Experts model with 256 experts (4 active per token), GatedDeltaNet SSM layers + full attention layers, and built-in vision capabilities.

It has been:

1. JANG quantized — JANG_4K profile (8-bit attention, 5-bit important, 3-bit experts) — 18 GB
2. CRACK abliterated — permanent weight-level removal of safety refusal

HarmBench Results

315/320 (98.4%) — tested with enable_thinking=false, temperature=1.0

JANG vs MLX Uniform Quantization

JANG's mixed-precision approach preserves knowledge better than MLX uniform quantization at the same size, and runs faster due to smaller memory footprint.

MMLU Results

65 curated hard questions across 13 subjects. Surgery preserves knowledge almost perfectly.

Safety guardrails were NOT helping with reasoning — removing them slightly improved math and physics scores.

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-35B-A3B-JANG_4K-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 (chain-of-thought reasoning before answering).

To disable thinking for faster responses:

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

Tip: Use temperature=1.0 for chat (greedy can cause repetition). Use temperature=0.0 for structured tasks like MMLU.

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 using per-layer projected vectors from structurally-mirrored prompt pairs.

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

Qwen 3.5 VL 35B — JANG_4K + CRACK

pip install "jang[mlx]"

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