HEP-BNN is a framework for finding Low-Latency execution configurations of BNNs on Heterogeneous multiprocessor platforms.

Presentation Paper on arXiv

First set up a conda environment with all the necessary dependencies:

git clone https://github.com/LeonardDavid/hep-bnn.git
cd fastinference
conda env create -f environment.yml
conda activate fi

Please note that this environment also contains some larger packages such as PyTorch so the installation may take some time.

Execute the following command before first time use, and after any changes to any of the Jinja2 files, or any implement.py files :

python setup.py install

Add your BNN model (in ONNX format) under

fastinference/implementations/neuralnet/cuda/automatic/model/

and the testing data (in csv format) under

fastinference/implementations/neuralnet/cuda/automatic/data/

Also see the example files for the Fashion-MNIST and CIFAR10 models, used during development.

Call the following command:

fastinference/implementations/neuralnet/cuda/automatic/test_cuda.py --outpath tmp/fastinference/cuda_auto --dataset cifar

Maintaner [email protected]

Special thanks goes to Sebastian Buschjäger for providing the original FastInference framework as a base.