syclnn

SYCL-accelerated feed-forward neural network library.

Uses Intel's DPC++ compiler and UXL Foundation's oneMath library.

Python bindings are provided.

Requirements

• CMake
• Intel® oneAPI Base Toolkit
• (optional, for NVIDIA GPUs) CUDA Toolkit and oneAPI for NVIDIA® GPUs

Installation and usage

pip install .

# for NVIDIA GPU support
pip install . --config-settings=cmake.args=-DENABLE_NVIDIA_GPU=ON

Then you can import the module in Python

Python 3.12.5 (main, Apr  2 2025, 00:00:00) [GCC 11.5.0 20240719 (Red Hat 11.5.0-5)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import syclnn
>>> dir(syclnn)
['ActivationType', 'AdaGrad', 'Adam', 'AdaptiveLearningRate_double', 'AdaptiveLearningRate_float', 'AdaptiveLearningStrategy_double', 'AdaptiveLearningStrategy_float', 'BackPropagation', 'Classical', 'Constant', 'Disabled', 'ELU', 'ElasticNet', 'L1', 'L2', 'LayerDescription', 'LeakyReLU', 'LinearDecay', 'MaxEpochs', 'MinError', 'MinErrorChange', 'MomentumConfig_double', 'MomentumConfig_float', 'MomentumType_double', 'MomentumType_float', 'Network_double', 'Network_float', 'RMSProp', 'ReLU', 'RegularizationType_double', 'RegularizationType_float', 'Regularization_double', 'Regularization_float', 'Sigmoid', 'Standard', 'StopCriteriaType_double', 'StopCriteriaType_float', 'StopCriteria_double', 'StopCriteria_float', 'Tanh', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', 'get_sycl_devices']

Using a container

The container uploaded to GitHub's registry contains a Python environment with the syclnn module already built and installed.

Running

docker run --rm --name syclnn-env -it ghcr.io/napaalm/syclnn:latest

# for NVIDIA GPU support
docker run --rm --gpus all --name syclnn-env -it ghcr.io/napaalm/syclnn:nvidia

Run python3.12 inside this container.

Running with Podman

podman run --rm --name syclnn-env -it syclnn:latest

# for NVIDIA GPU support
podman run --rm --device nvidia.com/gpu=all --security-opt=label=disable --name syclnn-env -it syclnn:nvidia

Build it yourself

docker build -t syclnn .

# for NVIDIA GPU support
docker build -t syclnn . --build-arg ENABLE_NVIDIA=1