Installing PyTorch on Jetson-TX2 for Python2.7

install_pytorch.sh

# First of all install python pip
sudo apt-get install python-pip

# Upgrading pip using `pip install -U pip` gave me errors after,
# so I will be avoiding that.

# Clone pytorch repo
git clone http://github.com/pytorch/pytorch

# Enter the folder
cd pytorch/

# Checkout to v0.4.0 (Version I want)
git checkout -b v0.4.0 v0.4.0

# Get the required submodules
git submodule update --init

# install prereqs
sudo pip install -U setuptools
sudo pip install -r requirements.txt

# Set environment variables
export CUDNN_LIB_DIR=/usr/lib/aarch64-linux-gnu
export CUDNN_INCLUDE_DIR=/usr/include
export CUDA_HOME=/usr/local/cuda  # Provide path where CUDA is present

# Install PyTorch
sudo python setup.py install  # Pytorch
sudo pip install torchvision  # Torchvision package that supports the PyTorch library

# Testing if PyTorch with GPU is installed
# import torch
# print(torch.__version__)
## Out: 0.4.0a0+3749c58
# print(torch.cuda.is_available())
## Out: True