First make sure docker is installed. The docker plugin nvidia-docker is required to run on Nvidia GPUs.
Pre-built docker containers are available at https://hub.docker.com/r/mxnet/
For example, the following command launches a container with the Python package installed. It will pull the docker images from docker hub if it does not exist locally.
docker run -ti --rm mxnet/python
Then you can run MXNet in python, e.g.:
# python -c 'import mxnet as mx; a = mx.nd.ones((2,3)); print((a*2).asnumpy())' [[ 2. 2. 2.] [ 2. 2. 2.]]
If the host machine has at least one GPU installed and nvidia-docker is installed, namely nvidia-docker run --rm nvidia/cuda nvidia-smi runs successfully, then you can run a container with GPU supports
nvidia-docker run -ti --rm mxnet/python:gpu
Now you can run the above example in GPU 0:
# python -c 'import mxnet as mx; a = mx.nd.ones((2,3), mx.gpu(0)); print((a*2).asnumpy())' [[ 2. 2. 2.] [ 2. 2. 2.]]
All images are based on Ubuntu 14.04. The gpu tag is built with CUDA 8.0 and cuDNN 5.
Hosted at https://hub.docker.com/r/mxnet/python/
Python versions: 2.7.12 and 3.5.2.
Available tags:
Hosted at https://hub.docker.com/r/mxnet/r-lang/
R version: 3.3.3
Available tags:
Hosted at https://hub.docker.com/r/mxnet/julia/
Julia version: 0.5.1
Available tags:
Hosted at https://hub.docker.com/r/mxnet/scala/
Scala version: 2.11.8
Available tags:
Hosted at https://hub.docker.com/r/mxnet/perl/
Perl version: 5.18.2
Available tags:
The following command build the default Python package
./tool.sh build python cpu
Run ./tool.sh for more details. Use
Tips: The following commands stop all docker containers and delete all docker images.
docker stop $(docker ps -a -q) docker rm $(docker ps -a -q)
docker rmi $(docker images -a -q)