Qwen3.6-27B-int4-mlx-AutoRound

Model Details

This model is an mlx format W4A16 quantized model with group_size 128 and symmetric quantization of Qwen/Qwen3.6-27B generated by intel/auto-round. Please follow the license of the original model.

We currently support this format in AutoRound, but do not have the hardware to validate this large model.

As a result, we are unable to verify whether it runs correctly or achieves expected performance.

We would greatly appreciate your help in testing it, and welcome any contributions to our open-source project.

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/Qwen3.6-27B-4.5b-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)

Generate the Model

ar>0.12.2

We also adopt --iters 1000 --nsamples 512. For a better resource trade-off, you could use the following command:

  AR_DISABLE_COPY_MTP_WEIGHTS=1 CUDA_VISIBLE_DEVICES=$device python3 -m auto_round \
  --scheme W4A16 \
  --format mlx \
  --output_dir "./test_mlx" \
  2>&1 | tee -a test_mlx.txt

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