apodex-1.0-0.8B-SFT-rebased-AutoRound-W4A16-Tuning

Model Details

This model is a int4 weight-only quantization with group_size 128 and symmetric quantization of dark-pen/apodex-1.0-0.8B-SFT-rebased generated by TUNING. Please follow the license of the original model.

Quantization Details

Attribute	Value
Base Model	dark-pen/apodex-1.0-0.8B-SFT-rebased
Quantization Tool	TUNING
Quantization Scheme	W4A16
Quantized Size	925 MB

Evaluation Results

Task	Accuracy
hellaswag	0.4030
mmlu	0.5018
mmlu_abstract_algebra	0.2900
mmlu_anatomy	0.4444
mmlu_astronomy	0.5329
mmlu_business_ethics	0.4800
mmlu_clinical_knowledge	0.5811
mmlu_college_biology	0.5972
mmlu_college_chemistry	0.4200
mmlu_college_computer_science	0.3900
mmlu_college_mathematics	0.4000
mmlu_college_medicine	0.5780
mmlu_college_physics	0.4020
mmlu_computer_security	0.6200
mmlu_conceptual_physics	0.5277
mmlu_econometrics	0.3596
mmlu_electrical_engineering	0.5517
mmlu_elementary_mathematics	0.3783
mmlu_formal_logic	0.4603
mmlu_global_facts	0.3000
mmlu_high_school_biology	0.6774
mmlu_high_school_chemistry	0.5665
mmlu_high_school_computer_science	0.5600
mmlu_high_school_european_history	0.5818
mmlu_high_school_geography	0.6768
mmlu_high_school_government_and_politics	0.6269
mmlu_high_school_macroeconomics	0.5410
mmlu_high_school_mathematics	0.3222
mmlu_high_school_microeconomics	0.5714
mmlu_high_school_physics	0.3775
mmlu_high_school_psychology	0.7119
mmlu_high_school_statistics	0.4120
mmlu_high_school_us_history	0.5098
mmlu_high_school_world_history	0.5949
mmlu_human_aging	0.5336
mmlu_human_sexuality	0.5725
mmlu_humanities	0.4306
mmlu_international_law	0.6942
mmlu_jurisprudence	0.6667
mmlu_logical_fallacies	0.5521
mmlu_machine_learning	0.4464
mmlu_management	0.7476
mmlu_marketing	0.7778
mmlu_medical_genetics	0.5400
mmlu_miscellaneous	0.5824
mmlu_moral_disputes	0.5694
mmlu_moral_scenarios	0.2447
mmlu_nutrition	0.6144
mmlu_other	0.5529
mmlu_philosophy	0.5531
mmlu_prehistory	0.5031
mmlu_professional_accounting	0.3936
mmlu_professional_law	0.3442
mmlu_professional_medicine	0.4669
mmlu_professional_psychology	0.4755
mmlu_public_relations	0.5545
mmlu_security_studies	0.5837
mmlu_social_sciences	0.5886
mmlu_sociology	0.7363
mmlu_stem	0.4729
mmlu_us_foreign_policy	0.6200
mmlu_virology	0.4337
mmlu_world_religions	0.5965
piqa	0.6953

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 = "apodex-1.0-0.8B-SFT-rebased-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 apodex-1.0-0.8B-SFT-rebased-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:

• Intel Neural Compressor
• AutoRound

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.