Qwen3.5-4B-AutoRound-MXFP4-RTN

Model Details

This model is a MXFP4 (Microscaling FP4) quantization of Qwen/Qwen3.5-4B generated by AutoRound. Please follow the license of the original model.

Quantization Details

Attribute	Value
Base Model	Qwen/Qwen3.5-4B
Quantization Tool	AutoRound
Quantization Scheme	MXFP4
Quantized Size	4300 MB

Evaluation Results

Task	Accuracy
hellaswag	0.5132
mmlu	0.6837
mmlu_abstract_algebra	0.5000
mmlu_anatomy	0.7037
mmlu_astronomy	0.8224
mmlu_business_ethics	0.7500
mmlu_clinical_knowledge	0.7472
mmlu_college_biology	0.8472
mmlu_college_chemistry	0.5000
mmlu_college_computer_science	0.6700
mmlu_college_mathematics	0.5300
mmlu_college_medicine	0.6879
mmlu_college_physics	0.5000
mmlu_computer_security	0.7600
mmlu_conceptual_physics	0.7489
mmlu_econometrics	0.5702
mmlu_electrical_engineering	0.7172
mmlu_elementary_mathematics	0.6270
mmlu_formal_logic	0.5238
mmlu_global_facts	0.3300
mmlu_high_school_biology	0.8516
mmlu_high_school_chemistry	0.7094
mmlu_high_school_computer_science	0.8000
mmlu_high_school_european_history	0.7939
mmlu_high_school_geography	0.8384
mmlu_high_school_government_and_politics	0.8912
mmlu_high_school_macroeconomics	0.7103
mmlu_high_school_mathematics	0.4296
mmlu_high_school_microeconomics	0.8487
mmlu_high_school_physics	0.5497
mmlu_high_school_psychology	0.8844
mmlu_high_school_statistics	0.6944
mmlu_high_school_us_history	0.8333
mmlu_high_school_world_history	0.8228
mmlu_human_aging	0.7265
mmlu_human_sexuality	0.8092
mmlu_humanities	0.5983
mmlu_international_law	0.7851
mmlu_jurisprudence	0.7685
mmlu_logical_fallacies	0.7975
mmlu_machine_learning	0.5268
mmlu_management	0.8350
mmlu_marketing	0.8974
mmlu_medical_genetics	0.8100
mmlu_miscellaneous	0.7957
mmlu_moral_disputes	0.7052
mmlu_moral_scenarios	0.3966
mmlu_nutrition	0.7647
mmlu_other	0.7277
mmlu_philosophy	0.7074
mmlu_prehistory	0.7407
mmlu_professional_accounting	0.5496
mmlu_professional_law	0.4896
mmlu_professional_medicine	0.7574
mmlu_professional_psychology	0.7206
mmlu_public_relations	0.6727
mmlu_security_studies	0.7429
mmlu_social_sciences	0.7871
mmlu_sociology	0.8607
mmlu_stem	0.6667
mmlu_us_foreign_policy	0.8200
mmlu_virology	0.4759
mmlu_world_religions	0.7895
piqa	0.7519

How to Use

HF Usage

Step 1: Install AutoRound

pip install auto-round

Step 2: Load and run the quantized model

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen3.5-4B-AutoRound-MXFP4-RTN"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")

# prepare the model input
prompt = "Write a quick sort algorithm."
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(**model_inputs, max_new_tokens=512)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]) :].tolist()

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print("content:", content)

VLLM Usage

vllm serve Qwen3.5-4B-AutoRound-MXFP4-RTN \
    --trust-remote-code \
    --dtype bfloat16 \
    --tensor_parallel_size 1

If you encounter any issues, feel free to open an issue on the AutoRound GitHub repo or provide feedback on the Low-Bit Open LLM Leaderboard.

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:

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

This model is part of the Intel Low-Bit Open LLM Leaderboard initiative.