This model is a int4 weight-only quantization with group_size 128 and symmetric quantization of NousResearch/Hermes-4.3-36B generated by AutoRound. Please follow the license of the original model.

Quantization Details

Attribute	Value
Base Model	NousResearch/Hermes-4.3-36B
Quantization Tool	AutoRound
Quantization Scheme	W4A16
Quantized Size	20157 MB

Evaluation Results

Task	Accuracy
hellaswag	0.6316
mmlu	0.8314
mmlu_abstract_algebra	0.7000
mmlu_anatomy	0.8148
mmlu_astronomy	0.9408
mmlu_business_ethics	0.8200
mmlu_clinical_knowledge	0.8642
mmlu_college_biology	0.9444
mmlu_college_chemistry	0.6300
mmlu_college_computer_science	0.7800
mmlu_college_mathematics	0.6900
mmlu_college_medicine	0.8266
mmlu_college_physics	0.7353
mmlu_computer_security	0.8400
mmlu_conceptual_physics	0.8936
mmlu_econometrics	0.7281
mmlu_electrical_engineering	0.8621
mmlu_elementary_mathematics	0.9180
mmlu_formal_logic	0.7698
mmlu_global_facts	0.6000
mmlu_high_school_biology	0.9613
mmlu_high_school_chemistry	0.8079
mmlu_high_school_computer_science	0.9500
mmlu_high_school_european_history	0.8727
mmlu_high_school_geography	0.9343
mmlu_high_school_government_and_politics	0.9793
mmlu_high_school_macroeconomics	0.9308
mmlu_high_school_mathematics	0.6778
mmlu_high_school_microeconomics	0.9622
mmlu_high_school_physics	0.7815
mmlu_high_school_psychology	0.9523
mmlu_high_school_statistics	0.8333
mmlu_high_school_us_history	0.9412
mmlu_high_school_world_history	0.9198
mmlu_human_aging	0.8117
mmlu_human_sexuality	0.9084
mmlu_humanities	0.7666
mmlu_international_law	0.9174
mmlu_jurisprudence	0.8611
mmlu_logical_fallacies	0.8957
mmlu_machine_learning	0.7589
mmlu_management	0.8738
mmlu_marketing	0.9316
mmlu_medical_genetics	0.9400
mmlu_miscellaneous	0.9208
mmlu_moral_disputes	0.8381
mmlu_moral_scenarios	0.6637
mmlu_nutrition	0.8627
mmlu_other	0.8494
mmlu_philosophy	0.8714
mmlu_prehistory	0.8827
mmlu_professional_accounting	0.7943
mmlu_professional_law	0.6623
mmlu_professional_medicine	0.9044
mmlu_professional_psychology	0.8693
mmlu_public_relations	0.8182
mmlu_security_studies	0.8571
mmlu_social_sciences	0.9087
mmlu_sociology	0.9104
mmlu_stem	0.8351
mmlu_us_foreign_policy	0.9400
mmlu_virology	0.5241
mmlu_world_religions	0.8713
piqa	0.8139

How to Use

HF Usage

Step 1: Install AutoRound

pip install auto-round

Step 2: Load and run the quantized model

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Hermes-4.3-36B-AutoRound-W4A16-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

bash

vllm serve Hermes-4.3-36B-AutoRound-W4A16-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.

LeaderboardModel1/Hermes-4.3-36B-AutoRound-W4A16-RTN

Hermes-4.3-36B-AutoRound-W4A16-RTN

Model Details