Quickstart (Feeling Lucky)

Assumes you have Docker and are using a Mac

1
2
docker load -i /Volumes/CAFFE2/c2.gpu.tutorial.0.7.0.tar
docker run -it -p 8888:8888 cc2.gpu.tutorial.0.7.0 sh -c "jupyter notebook --no-browser --ip 0.0.0.0 /caffe2/caffe2/python/tutorials"

Essentially you need to locate the tar file, whatever its name is and import it with docker load -i <path-to-image-tar-file> Windows users: you can just change “/Volumes” to “D:" or whatever the drive letter the USB was assigned and it should work.

Setup Docker

You’ll need Docker installed on your local PC. Skip ahead if you’ve done this already. To be sure run this command:

1
docker version

Mac

Windows

Get Caffe2 Docker Image

You have two ways to do this. If you’re running this from a USB stick, then continue, if not, jump to the online option below.

Local/USB: Import the Caffe2 Docker Image

This image is in a tar file on the USB stick. You can import it by using this command:

1
docker load -i <path-to-image-tar-file>

Online: Pull the Caffe2 Docker Image

For the latest Docker image using GPU support and optional dependencies like IPython & OpenCV (don’t bother on Windows - see Troubleshooting notes):

1
docker pull caffe2ai/caffe2 && docker run -it caffe2ai/caffe2:latest

Launch the Image with Jupyter Notebook

Once the loading of the image finishes, check your list of Docker images:

1
docker images

Assuming it’s there you can now launch it:

1
docker run -it -p 8888:8888 c2.gpu.tutorial.0.7.0 sh -c "jupyter notebook --no-browser --ip 0.0.0.0 /caffe2/caffe2/python/tutorials"

This will output a URL. You just need to copy the provided URL/token combo into your browser and you should see the folder with tutorials.

In some situations you can’t access the Jupyter server on your browser via 0.0.0.0 or localhost. You need to pull the Docker IP address (run docker-machine ip) and use that to access the Jupyter server. If this doesn’t work, check your computer’s IP address and try that. If that doesn’t work, kill the server, start docker-machine as mentioned in troubleshooting, check its IP, then start the Jupyter server and use the docker-machine IP.

Using Docker and GPUs

To enable the power of your GPU while using Docker, you will want to install NVIDIA’s nvidia-docker. Use nvidia-docker run ... instead of docker run....

Troubleshooting

Getting Docker to run after installation may take some prodding and setting up of the environment. Try this:

1
2
docker-machine restart default
eval $(docker-machine env default)

More info on this setup is found on the Caffe2 docs site in Install>OS>Docker and in Install>OS>Cloud. The Cloud page has info specific to forwarding Docker through your SSH tunnel to your cloud server.

Troubleshooting  
common_gpu.cc:42 Found an unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. I will set the available devices to be zero.
Solution This may be a Docker-specific error where you need to launch the images while passing in GPU device flags: sudo docker run -ti --device /dev/nvidia0:/dev/nvidia0 --device /dev/nvidiactl:/dev/nvidiactl --device /dev/nvidia-uvm:/dev/nvidia-uvm mydocker-repo/mytag /bin/bash. You will need to update those devices according to your hardware (however this should match a 1-GPU build) and you need to swap out mydocker-repo/mytag with the ID or the repo/tag of your Docker image.
HyperV is not available on Home editions. Please use Docker Toolbox. Docker for Windows only works on Professional versions of Windows.
Solution Install Docker Toolbox. Don’t worry, the Caffe2 images should still work for you!
An error occurred trying to connect… various errors just after installing Docker Toolbox…
Solution run docker-machine env default then follow the instructions… run each of the commands that setup the docker environment then try docker version and you shouldn’t see the errors again and will be able to docker pull caffe2ai/caffe2.

Edit on GitHub