Installing Dragon

Get the Latest Version

$ Clone the Dragon repository

git clone --recursive https://github.com/seetaresearch/Dragon.git

We will call the directory that you cloned Dragon as REPO_ROOT.

Besides, let’s call the REPO_ROOT/Dragon as DRAGON_ROOT.

Installation

Step 1: Install C++ Dependencies

$ Setup Python(2.x, or 3.x) Development Environment

sudo apt-get install libpython3-dev

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

sudo apt-get install libprotobuf-dev protobuf-compiler

Step 2: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 3: Compile Dragon

$ Install CMake

sudo apt-get install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake .. -DWITH_CUDA=OFF -DWITH_CUDNN=OFF
make install -j $(nproc)

Step 4: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

Step 1: Preparing GPU Prerequisites

$ Download and install CUDA

$ (Optional) Download and install CUDNN

Step 2: Install C++ Dependencies

$ Setup Python(2.x, or 3.x) Development Environment

sudo apt-get install libpython3-dev

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

sudo apt-get install libprotobuf-dev protobuf-compiler

Step 3: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 4: Configure DRAGON_ROOT/CMakeLists.txt

$ (Optional) Set CUDA_ARCH for different architectures, see ComputeCapability

$ GCC version(4.8+, 5.0-) should add -std=c++11 to CUDA_NVCC_FLAGS, if nullptr is not found.

Step 5: Compile Dragon

$ Install CMake

sudo apt-get install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake ..
make install -j $(nproc)

Step 6: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

Step 1: Install C++ Dependencies

$ Setup Python(2.x, or 3.x) Development Environment

sudo apt-get install libpython3-dev

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

sudo apt-get install libprotobuf-dev protobuf-compiler

Step 2: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 3: Build MPI

cd $REPO_ROOT/ThirdParty/mpi
bash ./build.sh 0
sudo cp ./bin/mpirun /usr/bin

Step 4: Compile Dragon

$ Install CMake

sudo apt-get install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake .. -DWITH_CUDA=OFF -DWITH_CUDNN=OFF -DWITH_MPI=ON
make install -j $(nproc)

Step 5: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

Step 1: Preparing GPU Prerequisites

$ Download and install CUDA

$ (Optional) Download and install CUDNN

$ (Optional) Download and install NCCL

Step 2: Install C++ Dependencies

$ Setup Python(2.x or 3.x) Development Environment

sudo apt-get install libpython3-dev

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

sudo apt-get install libprotobuf-dev protobuf-compiler

Step 3: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 4: Configure DRAGON_ROOT/CMakeLists.txt

$ (Optional) Set CUDA_ARCH for different architectures, see ComputeCapability

$ GCC version(4.8+, 5.0-) should add -std=c++11 to CUDA_NVCC_FLAGS, if nullptr is not found.

Step 5: Build MPI

cd $REPO_ROOT/ThirdParty/mpi
bash ./build.sh
sudo cp ./bin/mpirun /usr/bin

Step 6: Compile Dragon

$ Install CMake

sudo apt-get install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake .. -DWITH_MPI=ON -DWITH_NCCL=ON
make install -j $(nproc)

Step 7: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

Step 1: Download ThirdParty.zip and unzip it under the REPO_ROOT

ThirdParty_vc12_x64.zip (Google Protobuf 3.5 For VS2013)

ThirdParty_vc14_x64.zip (Google Protobuf 3.6 For VS2015)

ThirdParty_vc141_x64.zip (Google Protobuf 3.6 For VS2017)

Step 2: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 3: Compile and Install Dragon

cd $REPO_ROOT/Scripts
start build_windows_vc[version].bat

Note: Modify the bat script to disable CUDA following the linux platform.

Installation

Step 1: Preparing GPU Prerequisites

$ Download and install CUDA

$ (Optional) Download and install CUDNN

Step 2: Download ThirdParty.zip and unzip it under the REPO_ROOT

ThirdParty_vc12_x64.zip (Google Protobuf 3.5 For VS2013)

ThirdParty_vc14_x64.zip (Google Protobuf 3.6 For VS2015)

ThirdParty_vc141_x64.zip (Google Protobuf 3.6 For VS2017)

Step 3: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 4: Configure DRAGON_ROOT/CMakeLists.txt

$ (Optional) Set CUDA_ARCH for different architectures, see ComputeCapability

Step 5: Compile and Install Dragon

cd $REPO_ROOT/Scripts
start build_windows_vc[version].bat

Installation

Disabled, as the Mircosoft MPI is not CUDA-Aware.

Installation

Disabled, as the Mircosoft MPI is not CUDA-Aware.

Installation

Step 1: Install C++ Dependencies

$ Setup Python(2.x or 3.x) Development Environment

brew install python

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

brew install protobuf libomp

Step 2: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 3: Compile Dragon

$ Install CMake

brew install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake .. -DWITH_CUDA=OFF -DWITH_CUDNN=OFF
make install -j $(getconf _NPROCESSORS_ONLN)

Step 4: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

Step 1: Preparing GPU Prerequisites

$ Download and install CUDA

$ (Optional) Download and install CUDNN

Step 2: Install C++ Dependencies

$ Setup Python Development Environment

brew install python

Note: You can also use Anaconda, A powerful toolkit for Data Science.

$ Setup C++ Development Environment

brew install protobuf libomp

Step 3: Install Python Requirements

cd $DRAGON_ROOT/python
pip install -r requirements.txt

Step 4: Configure DRAGON_ROOT/CMakeLists.txt

$ (Optional) Set CUDA_ARCH for different architectures, see ComputeCapability

$ Please confirm your CLANG_VERSION is compatible with NVCC_VERSION, See CUDA Install Guide for OSX.

Step 5: Compile Dragon

$ Install CMake

brew install cmake

$ Make

cd $DRAGON_ROOT
mkdir build
cd build
cmake ..
make install -j $(getconf _NPROCESSORS_ONLN)

Step 6: Install Dragon

cd $DRAGON_ROOT/python
python setup.py install

Note: If you do not have permission, try as follows:

cd $DRAGON_ROOT/python
python setup.py install --user

Installation

OSX dose not support MPI. May be you can contribute one?

Installation

OSX dose not support MPI. May be you can contribute one?