Sunday, September 11, 2016

Install notes for Tensorflow

Installed CUDA 7.5 and CuDNN 5.1.3

Steps modified from https://gist.github.com/erikbern/78ba519b97b440e10640

# Need Java 1.8 for bazel, on 14.04 this is a big mess - 14.10 and up openjdk-8 is in the repos
wget http://ftp.br.debian.org/debian/pool/main/libj/libjpeg-turbo/libjpeg62-turbo_1.3.1-12_amd64.deb
wget http://ftp.br.debian.org/debian/pool/main/libj/libjpeg6b/libjpeg62_6b2-2_amd64.deb
sudo dpkg --install --recursive --auto-deconfigure libjpeg62-turbo_1.3.1-12_amd64.deb
sudo apt-get update

sudo apt-get install debian-keyring debian-archive-keyring

sudo sh -c 'echo "deb http://httpredir.debian.org/debian/ jessie-backports main" >> /etc/apt/sources.list.d/debian-jessie-backports.list'

sudo sh -c 'echo "Package: *" >> /etc/apt/preferences.d/debian-jessie-backports'

sudo sh -c 'echo "Pin: release o=Debian,a=jessie-backports" >> /etc/apt/preferences.d/debian-jessie-backports'

sudo sh -c 'echo "Pin-Priority: -200" >> /etc/apt/preferences.d/debian-jessie-backports'
sudo apt-get -t jessie-backports install openjdk-8-jre openjdk-8-jre-headless openjdk-8-jdk-headless openjdk-8-jdk

# Set both to Java 8
sudo update-alternatives --config java
sudo update-alternatives --config javac

# Now for bazel
git clone https://github.com/bazelbuild/bazel
cd bazel
git checkout tags/0.2.1
./compile.sh
sudo cp output/bazel /usr/bin

# Tensorflow
git clone --recurse-submodules https://github.com/tensorflow/tensorflow
git checkout r0.10
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64"
export CUDA_HOME=/usr/local/cuda
TF_UNOFFICIAL_SETTING=1 ./configure
bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
pip install wheel
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
pip install /tmp/tensorflow_pkg/tensorflow-0.10.0-cp27-none-linux_x86_64.whl

Saturday, September 10, 2016

Get CuDNN from command line

Check the Dockerfile https://github.com/NVIDIA/nvidia-docker/blob/master/centos-7/cuda/7.5/runtime/cudnn5/Dockerfile