Changelog History
-
v0.8.1 Changes
August 08, 2017๐ This incremental release provides support for cuDNN7.
-
v0.8.0 Changes
July 21, 2017๐ Caffe2 v0.8.0 Release Notes
July 21, 2017Major Changes
- Caffe2 Brew API : now you can easily brew a variety of Caffe2 models with this help of this API
- Predictor Exporter and Mobile Exporter : export model state to a database or for mobile use; see the MNIST tutorial and the predictor code for examples of this and brew.
- 16-bit FP : Caffe2 and the NVIDIA Volta platform have added 16-bit Floating Point Training Support
๐ณ Docker
๐ณ
docker pull caffe2ai/caffe2
-
v0.7.0 Changes
April 18, 2017๐ Caffe2 v0.7.0 Release Notes
Installation
๐ This build is confirmed for:
- Ubuntu 14.04
- Ubuntu 16.06
Required Dependencies
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
๐ Optional GPU Support
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.
๐ง NVIDIA's detailed instructions or if you're feeling lucky try the quick install set of commands below.โก๏ธ 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
Install cuDNN (all Ubuntu versions)
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
Optional Dependencies
Note
libgflags2
is for Ubuntu 14.04.libgflags-dev
is for Ubuntu 16.04.# for Ubuntu 14.04sudo apt-get install -y --no-install-recommends libgflags2 # for Ubuntu 16.04sudo apt-get install -y --no-install-recommends libgflags-dev # for both Ubuntu 14.04 and 16.04sudo 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 sudo pip install \ flask \ graphviz \ hypothesis \ jupyter \ matplotlib \ pydot python-nvd3 \ pyyaml \ requests \ scikit-image \ scipy \ setuptools \ tornado
๐ Clone & Build
git clone --recursive https://github.com/caffe2/caffe2.git && cd caffe2 make && cd build && sudo make install python -c 'from caffe2.python import core' 2\>/dev/null && echo "Success" || echo "Failure"
๐ Run this command below to test if your GPU build was a success. You will get a test output either way, but it will warn you at the top of the output if CPU was used instead along with other errors like missing libraries.
python -m caffe2.python.operator\_test.relu\_op\_test
Environment Variables
๐ These environment variables may assist you depending on your current configuration. When using the install instructions above on the AWS Deep Learning AMI you don't need to set these variables. However, our Docker scripts built on Ubuntu-14.04 or NVIDIA's CUDA images seem to benefit from having these set. If you ran into problems with the build tests above then these are good things to check. Echo them first and see what you have and possibly append or replace with these directories. Also visit the troubleshooting section below.
echo $PYTHONPATH# export PYTHONPATH=/usr/local:$PYTHONPATH# export PYTHONPATH=$PYTHONPATH:/home/ubuntu/caffe2/buildecho $LD\_LIBRARY\_PATH# export LD\_LIBRARY\_PATH=/usr/local/lib:$LD\_LIBRARY\_PATH
Setting Up Tutorials & Jupyter Server
๐ If you're running this all on a cloud computer, you probably won't have a UI or way to view the IPython notebooks by default. Typically, you would launch them locally with
ipython notebook
and you would see a localhost:8888 webpage pop up with the directory of notebooks running. The following example will show you how to launch the Jupyter server and connect to remotely via an SSH tunnel.๐ First configure your cloud server to accept port 8889, or whatever you want, but change the port in the following commands. On AWS you accomplish this by adding a rule to your server's security group allowing a TCP inbound on port 8889. Otherwise you would adjust iptables for this.
Next you launch the Juypter server.
jupyter notebook --no-browser --port=8889
Then create the SSH tunnel. This will pass the cloud server's Jupyter instance to your localhost 8888 port for you to use locally. The example below is templated after how you would connect AWS, where
your-public-cert.pem
is your own public certificate and[email protected]
is your login to your cloud server. You can easily grab this on AWS by going to Instances > Connect and copy the part afterssh
and swap that out in the command below.ssh -N -f -L localhost:8888:localhost:8889 -i "your-public-cert.pem" [email protected]
Troubleshooting
Python errors 0๏ธโฃ Python version ๐ฆ Solution ๐ Python environment ๐ฆ Solution pip version If you plan to use Python with Caffe2 then you need pip. Solution sudo apt-get install python-pip
and also try using pip2 instead of pip."AttributeError: 'module' object has no attribute 'MakeArgument'" Occurs when calling core.CreateOperator
๐ Solution ๐ | Building from source | | | --- | --- | | OS version | Caffe2 requires Ubuntu 14.04 or greater. | | git | While you can download the Caffe2 source code and submodules directly from GitHub as a zip, using git makes it much easier. | | Solution |
sudo apt-get install git
| | protobuf | You may experience an error related to protobuf during the make step. | | Solution | Make sure you've installed protobuf in both of these two ways:sudo apt-get install libprotobuf-dev protobuf-compiler && sudo pip install protobuf
| | libgflags2 error | This optional dependency is for Ubuntu 14.04. | | Solution | Useapt-get install libgflags-dev
for Ubuntu 16.04. |๐ | GPU Support | | | --- | --- | ๐ | GPU errors | Unsupported GPU or wrong version | ๐ง | Solution | You need to know the specific
deb
for your version of Linux. `sudo dpkg -i | ๐ | Build issues | Be warned that installing CUDA and cuDNN will increase the size of your build by about 4GB, so plan to have at least 12GB for your Ubuntu disk size. | -
v0.6.0 Changes
April 03, 2017๐ Caffe2 v0.6.0 Release Notes
Installation
๐ Note: the release archive will not include third_party submodules which you will need in order to build Caffe2. We've uploaded an archive with the source and the submodules and attached it to this release. See the bottom of this page for the link!
๐ This build is confirmed for:
- Ubuntu 14.04
- Ubuntu 16.06
Required Dependencies
sudo apt-get update sudo apt-get install python-dev python-pip git build-essential cmake libprotobuf-dev protobuf-compiler libgoogle-glog-dev sudo pip install numpy protobuf
๐ Optional GPU Support
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.
๐ง NVIDIA's detailed instructions or if you're feeling lucky try the quick install set of commands below.โก๏ธ 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
Install cuDNN (all Ubuntu versions)
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
Optional Dependencies
sudo apt-get install libgtest-dev libgflags2 libgflags-dev liblmdb-dev libleveldb-dev libsnappy-dev libopencv-dev libiomp-dev openmpi-bin openmpi-doc libopenmpi-dev python-pydot sudo pip install flask graphviz hypothesis jupyter matplotlib pydot python-nvd3 pyyaml requests scikit-image scipy setuptools tornado
- Note for Ubuntu 16.04
libgflags2
should be replaced withlibgflags-dev
.
๐ Clone & Build
git clone --recursive https://github.com/caffe2/caffe2.git && cd caffe2 make && cd build && sudo make install python -c 'from caffe2.python import core' 2\>/dev/null && echo "Success" || echo "Failure"
๐ Run this command below to test if your GPU build was a success. You will get a test output either way, but it will warn you at the top of the output if CPU was used instead along with other errors like missing libraries.
python -m caffe2.python.operator\_test.relu\_op\_test
Environment Variables
๐ These environment variables may assist you depending on your current configuration. When using the install instructions above on the AWS Deep Learning AMI you don't need to set these variables. However, our Docker scripts built on Ubuntu-14.04 or NVIDIA's CUDA images seem to benefit from having these set. If you ran into problems with the build tests above then these are good things to check. Echo them first and see what you have and possibly append or replace with these directories. Also visit the troubleshooting section below.
echo $PYTHONPATH# export PYTHONPATH=/usr/local:$PYTHONPATH# export PYTHONPATH=$PYTHONPATH:/home/ubuntu/caffe2/buildecho $LD\_LIBRARY\_PATH# export LD\_LIBRARY\_PATH=/usr/local/lib:$LD\_LIBRARY\_PATH
Setting Up Tutorials & Jupyter Server
๐ If you're running this all on a cloud computer, you probably won't have a UI or way to view the IPython notebooks by default. Typically, you would launch them locally with
ipython notebook
and you would see a localhost:8888 webpage pop up with the directory of notebooks running. The following example will show you how to launch the Jupyter server and connect to remotely via an SSH tunnel.๐ First configure your cloud server to accept port 8889, or whatever you want, but change the port in the following commands. On AWS you accomplish this by adding a rule to your server's security group allowing a TCP inbound on port 8889. Otherwise you would adjust iptables for this.
Next you launch the Juypter server.
jupyter notebook --no-browser --port=8889
Then create the SSH tunnel. This will pass the cloud server's Jupyter instance to your localhost 8888 port for you to use locally. The example below is templated after how you would connect AWS, where
your-public-cert.pem
is your own public certificate and[email protected]
is your login to your cloud server. You can easily grab this on AWS by going to Instances > Connect and copy the part afterssh
and swap that out in the command below.ssh -N -f -L localhost:8888:localhost:8889 -i "your-public-cert.pem" [email protected]
Troubleshooting
Python errors 0๏ธโฃ Python version ๐ฆ Solution ๐ Python environment ๐ฆ Solution pip version If you plan to use Python with Caffe2 then you need pip. Solution sudo apt-get install python-pip
and also try using pip2 instead of pip.๐ | Building from source | | | --- | --- | | OS version | Caffe2 requires Ubuntu 14.04 or greater. | | git | While you can download the Caffe2 source code and submodules directly from GitHub as a zip, using git makes it much easier. | | Solution |
sudo apt-get install git
| | protobuf | You may experience an error related to protobuf during the make step. | | Solution | Make sure you've installed protobuf in both of these two ways:sudo apt-get install libprotobuf-dev protobuf-compiler && sudo pip install protobuf
| | libgflags2 error | This optional dependency is for Ubuntu 14.04. | | Solution | Useapt-get install libgflags-dev
for Ubuntu 16.04. |๐ | GPU Support | | | --- | --- | ๐ | GPU errors | Unsupported GPU or wrong version | ๐ง | Solution | You need to know the specific
deb
for your version of Linux. `sudo dpkg -i | ๐ | Build issues | Be warned that installing CUDA and cuDNN will increase the size of your build by about 4GB, so plan to have at least 12GB for your Ubuntu disk size. |