dealign.ai

Qwen3.5-VL-0.8B-8bit-MLX-CRACK

CRACK (Controlled Residual Abliteration of Compliance Knowledge) applied to Qwen 3.5 VL 0.8B, quantized to 8-bit MLX format.

Key Specs

Property Value
Base Model Qwen3.5-VL-0.8B
Architecture 24 layers, Dense + VL
Method CRACK (multi-vector weight surgery)
Quantization 8-bit (9.389 bpw)
Speed ~275 tok/s (Apple Silicon)
Compliance 7/8 (88%)
Coherence UTR=0.71 (Good)

What is CRACK?

CRACK is a surgical weight modification technique that removes safety-trained refusal behaviors from language models while preserving core capabilities. Unlike prompt-based jailbreaks, CRACK modifies the model weights directly, making the changes permanent and robust.

⚠️ Warning: This model will comply with requests that the original model would refuse. Use responsibly.

Usage

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

model, processor = load("dealignai/Qwen3.5-VL-0.8B-8bit-MLX-CRACK")
config = load_config("dealignai/Qwen3.5-VL-0.8B-8bit-MLX-CRACK")

prompt = apply_chat_template(processor, config, "Your prompt here", num_images=0)
output = generate(model, processor, prompt=prompt, max_tokens=256)
print(output)

Qwen 3.5 CRACK Family

Model Quant Speed Compliance Access
0.8B Q4 303 tok/s 8/8 Free
0.8B Q8 275 tok/s 7/8 Free
2B Q4, Q8 248/187 tok/s 8/8 Free
4B Q4, Q8 8/8 Free
9B Q4, Q8 101 tok/s 8/8 Free
35B-A3B Q4, Q8 ~80 tok/s 8/8 Free
122B-A10B Q4, Q6, Q8 8/8 Gated
397B-A17B MLX, FP16 Gated

About dealignai

Dealign.AI Mascot

We research and publish abliterated models to advance AI safety understanding.

Follow us: 𝕏 @dealignai

License

Apache 2.0 (same as base model).


Support dealignai

All models are built from original research and published for free. These models are specifically crafted to be excellent coders and general-purpose assistants.

Support us on Ko-fi — check out the Ko-fi membership for early access and extras.

Have questions or need help with a specific model? DM us — we help for free most of the time.

Ko-fi | X @dealignai | dealign.ai

Downloads last month
162
Safetensors
Model size
0.3B params
Tensor type
BF16
·
U32
·
F32
·
MLX
Hardware compatibility
Log In to add your hardware

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support