Qwen3-0.6B-AutoRound-MXFP4-Tuning

Model Details

This model is a MXFP4 (Microscaling FP4) quantization of Qwen/Qwen3-0.6B generated by TUNING. Please follow the license of the original model.

Quantization Details

Attribute	Value
Base Model	Qwen/Qwen3-0.6B
Quantization Tool	TUNING
Quantization Scheme	MXFP4
Quantized Size	767 MB

Evaluation Results

Task	Accuracy
hellaswag	0.3469
mmlu	0.3444
mmlu_abstract_algebra	0.2700
mmlu_anatomy	0.3630
mmlu_astronomy	0.3224
mmlu_business_ethics	0.3600
mmlu_clinical_knowledge	0.3887
mmlu_college_biology	0.3889
mmlu_college_chemistry	0.3100
mmlu_college_computer_science	0.3200
mmlu_college_mathematics	0.2700
mmlu_college_medicine	0.3584
mmlu_college_physics	0.2451
mmlu_computer_security	0.4600
mmlu_conceptual_physics	0.3277
mmlu_econometrics	0.2632
mmlu_electrical_engineering	0.3655
mmlu_elementary_mathematics	0.3016
mmlu_formal_logic	0.2619
mmlu_global_facts	0.2200
mmlu_high_school_biology	0.3613
mmlu_high_school_chemistry	0.2709
mmlu_high_school_computer_science	0.3400
mmlu_high_school_european_history	0.4182
mmlu_high_school_geography	0.3384
mmlu_high_school_government_and_politics	0.3368
mmlu_high_school_macroeconomics	0.3154
mmlu_high_school_mathematics	0.2704
mmlu_high_school_microeconomics	0.3403
mmlu_high_school_physics	0.2649
mmlu_high_school_psychology	0.4495
mmlu_high_school_statistics	0.3380
mmlu_high_school_us_history	0.3873
mmlu_high_school_world_history	0.4346
mmlu_human_aging	0.4036
mmlu_human_sexuality	0.4275
mmlu_humanities	0.3158
mmlu_international_law	0.4132
mmlu_jurisprudence	0.3796
mmlu_logical_fallacies	0.3926
mmlu_machine_learning	0.2946
mmlu_management	0.4757
mmlu_marketing	0.5385
mmlu_medical_genetics	0.3900
mmlu_miscellaneous	0.3793
mmlu_moral_disputes	0.3237
mmlu_moral_scenarios	0.2380
mmlu_nutrition	0.3856
mmlu_other	0.3721
mmlu_philosophy	0.3376
mmlu_prehistory	0.3704
mmlu_professional_accounting	0.2624
mmlu_professional_law	0.2836
mmlu_professional_medicine	0.2647
mmlu_professional_psychology	0.3627
mmlu_public_relations	0.4000
mmlu_security_studies	0.4327
mmlu_social_sciences	0.3861
mmlu_sociology	0.5274
mmlu_stem	0.3191
mmlu_us_foreign_policy	0.4300
mmlu_virology	0.4096
mmlu_world_religions	0.3626
piqa	0.6289

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 = "Qwen3-0.6B-AutoRound-MXFP4-Tuning"

# 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 Qwen3-0.6B-AutoRound-MXFP4-Tuning \
    --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.