Model Details

This model is an int4 (W4A16, group_size 128, symmetric) model of mistralai/Ministral-3-8B-Instruct-2512 generated by intel/auto-round with 200 tuning iterations. Please follow the license of the original model.

  • Quantized (W4A16): attention, MLP
  • Kept in BF16: embeddings, lm_head, vision tower, multi-modal projector

How to Use

Environment

uv pip install vllm==0.32.0
uv pip install lm-eval==0.4.13

vLLM Usage

vllm serve INCModel4/Ministral-3-8B-Instruct-2512-W4A16-AutoRound \
    --tokenizer mistralai/Ministral-3-8B-Instruct-2512 \
    --dtype bfloat16

Evaluation

Benchmark BF16 W4A16 Recovery
GSM8K 86.81 86.05 99.13%
PIQA 79.49 79.11 99.52%
MMLU 73.06 72.19 98.81%
HellaSwag 58.64 58.03 98.95%
Average 74.50 73.84 99.12%

Evaluated with:

lm_eval \
    --model local-completions \
    --model_args model=INCModel4/Ministral-3-8B-Instruct-2512-W4A16-AutoRound,tokenizer=mistralai/Ministral-3-8B-Instruct-2512,tokenizer_backend=huggingface,base_url=http://localhost:8000/v1/completions,tokenized_requests=True,num_concurrent=32,max_retries=50,timeout=1800,max_length=131073,max_gen_toks=2048,add_bos_token=False \
    --tasks gsm8k \
    --batch_size 32 \
    --log_samples \
    --seed 42 \
    --num_fewshot 5 \
    --apply_chat_template \
    --fewshot_as_multiturn \
    --gen_kwargs max_gen_toks=2048

lm_eval \
    --model local-completions \
    --model_args model=INCModel4/Ministral-3-8B-Instruct-2512-W4A16-AutoRound,tokenizer=mistralai/Ministral-3-8B-Instruct-2512,tokenizer_backend=huggingface,base_url=http://localhost:8000/v1/completions,tokenized_requests=True,num_concurrent=32,max_retries=50,timeout=1800,max_length=131073,max_gen_toks=2048,add_bos_token=False \
    --tasks piqa,mmlu,hellaswag \
    --batch_size 32 \
    --log_samples \
    --seed 42

Generate the Model

auto-round \
    --model_name mistralai/Ministral-3-8B-Instruct-2512 \
    --scheme W4A16 \
    --iters 200 \
    --nsamples 128 \
    --ignore_layers vision_tower,multi_modal_projector,lm_head \
    --format auto_round \
    --output_dir Ministral-3-8B-Instruct-2512-W4A16-AutoRound \
    --seqlen 2048 \
    --batch_size 8 \
    --disable_trust_remote_code

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

Signed-off-by: Sun, Xuehao xuehao.sun@intel.com

Downloads last month
70
Safetensors
Model size
3B params
Tensor type
I32
·
BF16
·
F16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for INCModel4/Ministral-3-8B-Instruct-2512-W4A16-AutoRound

Paper for INCModel4/Ministral-3-8B-Instruct-2512-W4A16-AutoRound