These are the step by step instructions on how to install the Arm NN core, TensorflowLite Parser as well as PyArmNN for x86_64, Arm64 and Armhf for Ubuntu 20.04. The packages will also be added to Debian Bullseye, their progress can be tracked here: https://tracker.debian.org/pkg/armnn
sudo apt install software-properties-common sudo add-apt-repository ppa:armnn/ppa sudo apt update
We provide a number of packages for each architecture; x86_64, aarch64 and armhf as outlined below.
ARMNN_MAJOR_VERSION: This is the ABI version of the Arm NN source that has been packaged based on include/armnn/Version.hpp.
ARMNN_RELEASE_VERSION: This is the marketing release version based on the date source was released on github e.g. 20.11.
PACKAGE_VERSION: This is the version of the source package used to build the binaries packages from.
libarmnn-cpuref-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
libarmnntfliteparser{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
libarmnn{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
python3-pyarmnn_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
libarmnn-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
libarmnntfliteparser-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_amd64.deb
libarmnn-aclcommon{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn-cpuacc-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn-cpuref-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn-gpuacc-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnntfliteparser{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
python3-pyarmnn_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnntfliteparser-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_arm64.deb
libarmnn-aclcommon{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnn-cpuacc-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnn-cpuref-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnn-gpuacc-backend{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnntfliteparser{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnn{ARMNN_MAJOR_VERSION}_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
python3-pyarmnn_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnn-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
libarmnntfliteparser-dev_{ARMNN_RELEASE_VERSION}-{PACKAGE_VERSION}_armhf.deb
Due to Debian Packaging requiring the pristine tarball from our Github release, the version on Launchpad may not align with the released version on Github depending on the complexity of newly added features. In order to check for the latest available Arm NN version use apt-cache search:
apt-cache search libarmnn # This returns a list of matching packages, the latest being libarmnn23 i.e. ARMNN_MAJOR_VERSION=23 libarmnn-cpuref-backend23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-cpuref-backend24 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-dev - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnntfliteparser-dev - Arm NN is an inference engine for CPUs, GPUs and NPUs # Note: removal of dash to suit debian naming conventions libarmnn-tfliteparser23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnntfliteparser24 - Arm NN is an inference engine for CPUs, GPUs and NPUs # Note: removal of dash to suit debian naming conventions libarmnn23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn24 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-aclcommon23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnnaclcommon24 - Arm NN is an inference engine for CPUs, GPUs and NPUs # Note: removal of dash to suit debian naming conventions libarmnn-cpuacc-backend23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-cpuacc-backend24 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-gpuacc-backend23 - Arm NN is an inference engine for CPUs, GPUs and NPUs libarmnn-gpuacc-backend24 - Arm NN is an inference engine for CPUs, GPUs and NPUs # Export the ARMNN_MAJOR_VERSION to allow installation using the below examples export ARMNN_MAJOR_VERSION=24
The easiest way to install all of the available packages for your systems architecture is to run the command:
sudo apt-get install -y python3-pyarmnn libarmnn-cpuacc-backend${ARMNN_MAJOR_VERSION} libarmnn-gpuacc-backend${ARMNN_MAJOR_VERSION} libarmnn-cpuref-backend${ARMNN_MAJOR_VERSION}
# Verify installation via python:
python3 -c "import pyarmnn as ann;print(ann.GetVersion())"
# Returns '{ARMNN_MAJOR_VERSION}.0.0' e.g. 24.0.0
This will install PyArmNN and the three backends for Neon, Compute Library and our Reference Backend. It will also install their dependencies including the arm-compute-library package along with the Tensorflow Lite Parser and it's dependency Arm NN Core. If the user does not wish to use PyArmNN they can go up a level of dependencies and instead just install the Tensorflow Lite Parser:
sudo apt-get install -y libarmnntfliteparser${ARMNN_MAJOR_VERSION} libarmnn-gpuacc-backend${ARMNN_MAJOR_VERSION}
The easiest way to uninstall all of the previously installed packages is to run the command:
sudo apt autoremove -y libarmnn${ARMNN_MAJOR_VERSION}