The other day, there was a person who made caffe available on TSUBAME. (reference: https://twitter.com/sacred_fox/status/605297481765691392 https://twitter.com/sacred_fox/status/605297652331212800 http://tsubame.gsic.titech.ac.jp/labs)
When using with python, basically
~> source /usr/apps.sp3/nosupport/gsic/env/caffe-0.11.0.sh
Then import caffe.
With this script /usr/apps.sp3/nosupport/gsic/python/2.7.7/bin/ It will run the included python. However, if you have python installed in addition to the standard python because the version of python that is included in TSUBAME as standard is old and you want to use it, copy this script to an appropriate location and copy it as follows. Some comments are made out and used.
# Caffe: 0.11.0
# CUDA: 6.5
# HDF: 1.8.11
# Python: 2.7.8
# Boost: 1.58.0
source /usr/apps.sp3/env/set_cuda-6.5.sh
source /usr/apps.sp3/nosupport/gsic/env/all/opencv-3.0.0-rc1.sh
source /usr/apps.sp3/nosupport/gsic/env/all/boost-1.58.0.sh
#source /usr/apps.sp3/nosupport/gsic/env/all/python-2.7.7.sh
export PATH=/usr/apps.sp3/nosupport/gsic/caffe/0.11.0/bin/:$PATH
export LD_LIBRARY_PATH=/usr/apps.sp3/nosupport/gsic/caffe/0.11.0/lib/:/usr/apps.sp3/nosupport/hdf5/1.8.11/lib/:$LD_LIBRARY_PATH
export PYTHONPATH=/usr/apps.sp3/nosupport/gsic/caffe/0.11.0/python/:$PYTHONPATH
export CPATH=/usr/apps.sp3/nosupport/gsic/caffe/0.11.0/include/:/usr/apps.sp3/cuda/6.5/include/:/usr/apps.sp3/isv/intel/xe2013.1.046/mkl/include/:/usr/apps.sp3/nosupport/hdf5/1.8.11/include/:$CPATH
export LIBRARY_PATH=/usr/apps.sp3/nosupport/gsic/caffe/0.11.0/lib/:/usr/apps.sp3/nosupport/hdf5/1.8.11/lib/:$LIBRARY_PATH
In this case, of course, it will not work unless you install caffe's dependent library in your python environment. The dependencies are as described on the official caffe page (http://caffe.berkeleyvision.org/), and many can be installed with pip. However, since only protobuf cannot be installed with pip, it is necessary to drop the source code from the official and execute the following command in the installed directory.
~> ./configure
~> make
~> make install
~> cd python
~> python setup.py build
~> python setup.py test
~> python setup.py install
Recommended Posts