Intel/Qwen3.5-4B-int4-mlx-AutoRound

Model Details

This model is an mlx format int4 model with group_size 128 and symmetric quantization of Qwen/Qwen3.5-4B generated by intel/auto-round. Please follow the license of the original model.

MLX-VLM inference

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

model_name_or_path= "Intel/Qwen3.5-4B-int4-mlx-AutoRound"

model, processor = load(model_name_or_path)
mlx_cfg = load_config(model_name_or_path)
prompt_text = "Describe this image in one sentence."
formatted = apply_chat_template(processor, mlx_cfg, prompt_text, num_images=1)
# Use a public example image so the test does not need local assets.
image_url = "https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3.5/demo/RealWorld/RealWorld-04.png"

output = generate(model, processor, formatted, image=[image_url], max_tokens=2048).text
print(output)

"""
The user wants a one-sentence description of the image.

1.  **Identify the main subject:** A large, stone statue of a head with a golden headband is in the foreground.
2.  **Identify the background:** Behind the statue is a sprawling city built on a hillside, with dense housing and some taller buildings, set against a backdrop of mountains and a blue sky with clouds.
3.  **Identify foreground details:** There's a sign that says "Origen" on a railing in the immediate foreground.
4.  **Synthesize into a single sentence:**
    *   *Draft 1:* A large stone statue with a gold headband stands in the foreground overlooking a dense city built on a hillside with mountains in the background.
    *   *Draft 2:* A large stone statue of a head wearing a golden headband overlooks a sprawling, densely packed city nestled on a hillside, with mountains visible in the distance.
    *   *Refining for flow and detail:* A large stone statue of a head adorned with a golden headband stands in the foreground overlooking a sprawling, densely packed city built on a hillside, with mountains and a blue sky visible in the background.

Let's try to make it punchier.
"A large stone statue of a head with a golden headband overlooks a sprawling, densely packed city built on a hillside, with mountains and a blue sky in the background."

This covers the main elements: the statue, the city, the terrain, and the sky.

Let's double check the "Origen" sign. It's a detail, but maybe not essential for a *general* one-sentence description unless I want to be very specific. I'll stick to the main visual components.

Revised sentence: A large stone statue of a head adorned with a golden headband overlooks a sprawling, densely packed city built on a hillside, with mountains and a blue sky visible in the background.

Wait, looking closer at the statue, it looks like a stylized face, maybe indigenous or mythological. "Stone statue of a head" is accurate. The background is definitely a city on a hill.

Final Polish: A large stone statue of a head adorned with a golden headband overlooks a sprawling, densely packed city built on a hillside, with mountains and a blue sky visible in the background.
</think>

A large stone statue of a head adorned with a golden headband overlooks a sprawling, densely packed city built on a hillside, with mountains and a blue sky visible in the background.
"""

Generate the Model

this pr is required https://github.com/intel/auto-round/pull/1732

AR_DISABLE_COPY_MTP_WEIGHTS=1 auto-round \
  --model Qwen/Qwen3.5-4B \
  --format mlx \
  --output_dir "Qwen3.5-4B-int4"

Ethical Considerations and Limitations

The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. Because of the limitations of the pretrained model and the finetuning datasets, it is possible that this model could generate lewd, biased or otherwise offensive outputs.

Therefore, before deploying any applications of the model, developers should perform safety testing.

Caveats and Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model.

Here are a couple of useful links to learn more about Intel's AI software:

• Intel Neural Compressor

Disclaimer

The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please consult an attorney before using this model for commercial purposes.

Cite

@article{cheng2023optimize, title={Optimize weight rounding via signed gradient descent for the quantization of llms}, author={Cheng, Wenhua and Zhang, Weiwei and Shen, Haihao and Cai, Yiyang and He, Xin and Lv, Kaokao and Liu, Yi}, journal={arXiv preprint arXiv:2309.05516}, year={2023} }

arxiv github