Install CUDA 9.2, cuDNN 7.2.1, Anaconda and PyTorch on Ubuntu 16.04.

pytorch_setup.sh

# tested on AWS p2.xlarge August 29, 2018

# install CUDA
sudo apt-get update && sudo apt-get install wget -y --no-install-recommends
CUDA_URL="https://developer.nvidia.com/compute/cuda/9.2/Prod2/local_installers/cuda-repo-ubuntu1604-9-2-local_9.2.148-1_amd64"
wget -c ${CUDA_URL} -O cuda.deb
sudo dpkg --install cuda.deb
sudo apt-key add /var/cuda-repo-9-2-local/7fa2af80.pub
sudo apt-get update
sudo apt-get install -y cuda
rm cuda.deb

CUDA_PATCH_URL="https://developer.nvidia.com/compute/cuda/9.2/Prod2/patches/1/cuda-repo-ubuntu1604-9-2-148-local-patch-1_1.0-1_amd64"
wget -c ${CUDA_PATCH_URL} -O cuda_patch.deb
sudo dpkg --install cuda_patch.deb
sudo apt-get upgrade -y cuda
rm cuda_patch.deb

# install cuDNN 7.2.1
CUDNN_URL="http://developer.download.nvidia.com/compute/redist/cudnn/v7.2.1/cudnn-9.2-linux-x64-v7.2.1.38.tgz"
wget -c ${CUDNN_URL} -O cudnn.tgz
sudo tar -xzf cudnn.tgz -C /usr/local
rm cudnn.tgz
sudo ldconfig

# install nccl
wget http://file.lzhu.me/cuda/nccl-repo-ubuntu1604-2.3.5-ga-cuda9.0_1-1_amd64.deb
sudo dpkg -i nccl-repo-ubuntu1604-2.3.5-ga-cuda9.0_1-1_amd64.deb
sudo apt-get update
sudo apt-get --assume-yes install libnccl2=2.3.5-2+cuda9.0 libnccl-dev=2.3.5-2+cuda9.0


# add variables to .profile
echo 'export CUDA_HOME=/usr/local/cuda' >> ~/.profile
echo 'export PATH=${CUDA_HOME}/bin:${PATH}' >> ~/.profile
echo 'export LD_LIBRARY_PATH=${CUDA_HOME}/lib64:$LD_LIBRARY_PATH' >> ~/.profile

# install conda
wget -c "https://repo.anaconda.com/archive/Anaconda3-5.2.0-Linux-x86_64.sh" -O conda.sh
bash conda.sh -b
rm conda.sh
echo 'export PATH=${HOME}/anaconda3/bin:${PATH}' >> ~/.profile
echo "source ~/.profile" >> ~/.bashrc

# setup environment variables
source ~/.profile

# install pytorch
conda install -y pytorch torchvision cuda92 -c pytorch