| commit | 1e5140aa76bac9675b2ea560f85680c33bd2a1e7 | [log] [tgz] |
|---|---|---|
| author | Aapo Kyrola <akyrola@fb.com> | Tue Apr 11 13:00:44 2017 -0700 |
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | Tue Apr 11 13:03:48 2017 -0700 |
| tree | a7f278b3dfcd9f60b5a62dfcbbf59371a51d748c | |
| parent | 15c6f637d6a7eef74aaa11a5cccb1115f96e9a5d [diff] |
option to recompute blobs backward pass with massive memory savings Summary: This diff adds an option to recurrent_net to define some cell blobs to be recomputed on backward step, and thus they don't need to be stored in the step workspace. This is done by modifying the backward step to automatically include all operators that are needed to produce the output that is to be recomputed, and by storing those blobs in a shared workspace. To enable the shared workspace, i had to modify the stepworkspaces blob to also store a forward shared workspace. Making it a class field won't work since the lifecycle of the blob does not match the lifecycle of the operator. For basic LSTM, the performance hit is quite modest (about 15% with one setting, but your mileage might vary. For Attention models, I am sure this is beneficial as computing the attention blobs is not expensive. For basic LSTM, the memory saving is wonderful: each forward workspace only has 4 bytes (for timestep). I also modified the neural_mt LSTM Cells, but there is no test available, so I am not 100% sure I did it correctly. Please have a look. Added options to LSTM, MILSTM and LSTMAttention to enable memory mode. Reviewed By: urikz Differential Revision: D4853890 fbshipit-source-id: d8d0e0e75a5330d174fbfa39b96d8e4e8c446baa
Caffe2 is a deep learning framework made with expression, speed, and modularity in mind. It is an experimental refactoring of Caffe, and allows a more flexible way to organize computation.
Caffe2 is released under the BSD 2-Clause license.
Detailed build matrix (hit refresh if you see icons not showing up due to heroku):
| Target | Status |
|---|---|
| Linux | |
| Mac (CPU) | |
| Android | |
| iOS | |
| Linux + MKL | |
| Windows |
git clone --recursive https://github.com/caffe2/caffe2.git cd caffe2
brew install automake protobuf mkdir build && cd build cmake .. make
This build is confirmed for:
sudo apt-get update sudo apt-get install -y --no-install-recommends \ build-essential \ cmake \ git \ libgoogle-glog-dev \ libprotobuf-dev \ protobuf-compiler \ python-dev \ python-pip sudo pip install numpy protobuf
If you plan to use GPU instead of CPU only, then you should install NVIDIA CUDA and cuDNN, a GPU-accelerated library of primitives for deep neural networks. NVIDIA's detailed instructions or if you're feeling lucky try the quick install set of commands below.
Update your graphics card drivers first! Otherwise you may suffer from a wide range of difficult to diagnose errors.
For Ubuntu 14.04
sudo apt-get update && sudo apt-get install wget -y --no-install-recommends wget "http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_8.0.61-1_amd64.deb" sudo dpkg -i cuda-repo-ubuntu1404_8.0.61-1_amd64.deb sudo apt-get update sudo apt-get install cuda
For Ubuntu 16.04
sudo apt-get update && sudo apt-get install wget -y --no-install-recommends wget "http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_8.0.61-1_amd64.deb" sudo dpkg -i cuda-repo-ubuntu1604_8.0.61-1_amd64.deb sudo apt-get update sudo apt-get install cuda
CUDNN_URL="http://developer.download.nvidia.com/compute/redist/cudnn/v5.1/cudnn-8.0-linux-x64-v5.1.tgz"
wget ${CUDNN_URL}
sudo tar -xzf cudnn-8.0-linux-x64-v5.1.tgz -C /usr/local
rm cudnn-8.0-linux-x64-v5.1.tgz && sudo ldconfig
Note
libgflags2is for Ubuntu 14.04.libgflags-devis for Ubuntu 16.04.
# for Ubuntu 14.04 sudo apt-get install -y --no-install-recommends libgflags2
# for Ubuntu 16.04 sudo apt-get install -y --no-install-recommends libgflags-dev
# for both Ubuntu 14.04 and 16.04 sudo apt-get install -y --no-install-recommends \ libgtest-dev \ libiomp-dev \ libleveldb-dev \ liblmdb-dev \ libopencv-dev \ libopenmpi-dev \ libsnappy-dev \ openmpi-bin \ openmpi-doc \ python-pydot
Check the Python section below and install optional packages before you build.
mkdir build && cd build cmake .. make
We use CMake's Android and iOS ports to build native binaries that you can then integrate into your Android or XCode projects. See scripts/build_android.sh and scripts/build_ios.sh for more details.
For Android, one can also use gradle to build Caffe2 directly with Android Studio. An example project can be found here. Note that you may need to configure Android Studio so that it has the right SDK and NDK versions to build the code.
For Raspbian, run scripts/build_raspbian.sh on the Raspberry Pi.
To install Caffe2 on NVidia's Tegra X1 platform, simply install the latest system with the NVidia JetPack installer, and then run scripts/build_tegra_x1.sh on the Tegra device.
To run the tutorials you'll need ipython-notebooks and matplotlib, which can be installed on OS X with:
brew install matplotlib --with-python3 pip install ipython notebook
You may also find these required for specific tutorials and examples, so you can run this to get all of the prerequisites at once:
sudo pip install \
flask \
graphviz \
hypothesis \
jupyter \
matplotlib \
pydot python-nvd3 \
pyyaml \
requests \
scikit-image \
scipy \
setuptools \
tornado
Ubuntu 14.04 (GCC)
OS X (Clang)
Options (both Clang and GCC)
BLAS
Other