chainer

2017/5/26

環境 †

python = 3.6.0 cuda = 8.0

手順 †

% git clone https://github.com/pfnet/chainer.git
% export LD_LIBRARY_PATH=/Developer/NVIDIA/CUDA-8.0/lib:$LD_LIBRARY_PATH

(とりあえず)

cudnnのライブラリをダウンロードする。https://developer.nvidia.com/cudnn (requires membership of the NVIDIA Developer Program)

% tar -xvf cudnn-8.0-osx-x64-v6.0.tar
% sudo cp cuda/lib/libcudnn* /Developer/NVIDIA/CUDA-8.0/lib/
% sudo cp cuda/include/cudnn.h /Developer/NVIDIA/CUDA-8.0/include/

% cd  /Developer/NVIDIA/CUDA-8.0/bin
% sh cuda-install-samples-8.0.sh ~/workspace/chainer/

% pip3 install chainer
% python3
    >>> import cupy
    >>> import cupy.cudnn

% cd workspace/chainer/NVIDIA_CUDA-8.0_Samples
% make
% cd bin/x86_64/darwin/release
% ./smokeParticles

動けば成功

% cd chainer
% pip3 install -U setuptools
% python setup.py install
% pip3 install cython 
% pip3 install matplotlib
% cd examples/mnist/
% python3 train_mnist.py -g 0

デバッグ †

% pip3 install pdb
% python3 -m pdb train_mnist.py -g 0