Supra-50M-Reasoning-AutoRound-W4A16-Tuning

Model Details

This model is a int4 weight-only quantization with group_size 128 and symmetric quantization of SupraLabs/Supra-50M-Reasoning generated by TUNING. Please follow the license of the original model.

Quantization Details

Attribute	Value
Base Model	SupraLabs/Supra-50M-Reasoning
Quantization Tool	TUNING
Quantization Scheme	W4A16
Quantized Size	49 MB

Evaluation Results

Task	Accuracy
hellaswag	0.2771
mmlu	0.2347
mmlu_abstract_algebra	0.2100
mmlu_anatomy	0.2815
mmlu_astronomy	0.1974
mmlu_business_ethics	0.2800
mmlu_clinical_knowledge	0.2491
mmlu_college_biology	0.2986
mmlu_college_chemistry	0.1700
mmlu_college_computer_science	0.2700
mmlu_college_mathematics	0.2200
mmlu_college_medicine	0.2139
mmlu_college_physics	0.2353
mmlu_computer_security	0.2800
mmlu_conceptual_physics	0.2638
mmlu_econometrics	0.2456
mmlu_electrical_engineering	0.2414
mmlu_elementary_mathematics	0.1958
mmlu_formal_logic	0.2698
mmlu_global_facts	0.1700
mmlu_high_school_biology	0.2129
mmlu_high_school_chemistry	0.1970
mmlu_high_school_computer_science	0.2700
mmlu_high_school_european_history	0.2242
mmlu_high_school_geography	0.2222
mmlu_high_school_government_and_politics	0.1969
mmlu_high_school_macroeconomics	0.2128
mmlu_high_school_mathematics	0.2185
mmlu_high_school_microeconomics	0.2143
mmlu_high_school_physics	0.1788
mmlu_high_school_psychology	0.2000
mmlu_high_school_statistics	0.1574
mmlu_high_school_us_history	0.2745
mmlu_high_school_world_history	0.2743
mmlu_human_aging	0.3274
mmlu_human_sexuality	0.2290
mmlu_humanities	0.2414
mmlu_international_law	0.2149
mmlu_jurisprudence	0.2685
mmlu_logical_fallacies	0.2270
mmlu_machine_learning	0.3036
mmlu_management	0.1748
mmlu_marketing	0.2863
mmlu_medical_genetics	0.3200
mmlu_miscellaneous	0.2529
mmlu_moral_disputes	0.2543
mmlu_moral_scenarios	0.2346
mmlu_nutrition	0.2288
mmlu_other	0.2475
mmlu_philosophy	0.1768
mmlu_prehistory	0.2006
mmlu_professional_accounting	0.2376
mmlu_professional_law	0.2471
mmlu_professional_medicine	0.1765
mmlu_professional_psychology	0.2533
mmlu_public_relations	0.2091
mmlu_security_studies	0.2000
mmlu_social_sciences	0.2220
mmlu_sociology	0.2338
mmlu_stem	0.2245
mmlu_us_foreign_policy	0.2600
mmlu_virology	0.2892
mmlu_world_religions	0.3216
piqa	0.6012

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 = "Supra-50M-Reasoning-AutoRound-W4A16-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 Supra-50M-Reasoning-AutoRound-W4A16-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.