Install PyTorch on Raspberry Pi 4

PyTorch is a crucial library to enable deep learning models to process in Raspberry Pi.

Procedure for Installing PyTorch and PyTorch Vision on Raspberry Pi 4B

1. First, update the Raspbian OS:

• sudo apt-get update
• sudo apt-get dist-upgrade

2. Install dependencies:

• sudo apt install libopenblas-dev libblas-dev m4 cmake cython python3-dev python3-yaml python3-setuptools
• sudo apt-get install libavutil-dev libavcodec-dev libavformat-dev libswscale-dev

3. You can install PyTorch and PyTorch Vision using the pre-compiled Python wheel files for saving your time for compiling:

• download wheel files
• sudo pip3 install torch-1.7.0a0-cp37-cp37m-linux_aarch64.whl 
• sudo pip3 install torchvision-0.8.0a0+45f960c-cp37-cp37m-linux_aarch64.whl 

4. Or, if you would like to get the different version by building PyTorch from the source yourself, which may take 6~10 hours (otherwise skip this step):

• For preventing OOM memory error, add swap memory:

  • sudo dphys-swapfile swapoff
  • sudo fallocate -l 2G /swapfile
  • sudo chmod 600 /swapfile
  • sudo mkswap /swapfile
  • suso swapon /swapfile
  • sudo swapon /swapfile
  • sudo nano /etc/fstab
  • sudo swapon --show
  • sudo free -h

• and for preventing blocked for more than 120 seconds. stack, run:

  • sudo sysctl -w vm.dirty_ratio=10
  • sudo sysctl -w vm.dirty_background_ratio=5

  Then commit your changes:

  • sudo sysctl -p

Compile Pytorch and Torchvision

1. Download source code and compile

• mkdir pytorch_installation
• cd pytorch_installation
• git clone --recursive https://github.com/pytorch/pytorch
• cd pytorch
• git checkout v1.6.0 # choose the version
• git submodule sync
• git submodule update --init --recursive
• git submodule update --remote third_party/protobuf

2. Edit CMakeLists.txt to set two flags:

• option(USE_PYTORCH_QNNPACK "Use ATen/QNNPACK (quantized 8-bit operators)" OFF) # By dedault is ON
• option(USE_SYSTEM_SLEEF "Use system-provided sleef." ON) # By default is OFF

3. Compile Pytorch

• export BUILD_TEST=0
• export USE_SYSTEM_NCCL=0
• export USE_CUDA=0
• export USE_MKLDNN=0
• export USE_NNPACK=0
• export USE_QNNPACK=0
• export USE_NUMPY=1
• export USE_DISTRIBUTED=0
• export NO_CUDA=1
• export NO_DISTRIBUTED=1
• export NO_MKLDNN=1
• export NO_NNPACK=1
• export NO_QNNPACK=1
• export ONNX_ML=1
• python3 setup.py build
• sudo -E python3 setup.py install
• python setup.py bdist_wheel # build whl file. The wheel will be in the dist/ folder.

4. Compile Torchvision

• cd ..
• mkdir pytorch_vision && cd pytorch_vision
• git clone --recursive https://github.com/pytorch/vision
• cd vision
• sudo -E python3 setup.py install
• python setup.py bdist_wheel # build whl file. The wheel will be in the dist/ folder.

5. Check the installation by running a simple test and confirming that Python outputs a tensor

• cd
• python3
• from __future__ import print_function
• import torch
• a = torch.rand(5,3)
• print (a)