Deep learning and AI frameworks for the Azure Data Science VM

Deep learning frameworks on the DSVM are listed below.

Caffe

Version(s) supported
Supported DSVM editions Linux (Ubuntu)
How is it configured / installed on the DSVM? Caffe is installed in /opt/caffe. Samples are in /opt/caffe/examples.
How to run it use X2Go to sign in to your VM, and then start a new terminal and enter the following:
cd /opt/caffe/examples
source activate root
jupyter notebook

A new browser window opens with sample notebooks. Binaries are installed in /opt/caffe/build/install/bin.

Installed version of Caffe requires Python 2.7 and won't work with Python 3.5, which is activated by default. To switch to Python 2.7, run source activate root to switch to Anaconda environment.

Caffe2

Version(s) supported
Supported DSVM editions Linux (Ubuntu)
How is it configured / installed on the DSVM? Caffe2 is installed in the [Python 2.7 (root) conda environment.
How to run it Terminal: Start Python, and import Caffe2.
* JupyterHub: Connect to JupyterHub, and then go to the Caffe2 directory to find sample notebooks. Some notebooks require the Caffe2 root to be set in the Python code; enter /opt/caffe2.

Chainer

Version(s) supported 5.2
Supported DSVM editions Linux (Ubuntu)
How is it configured / installed on the DSVM? Chainer is installed in Python 3.5.
How to run it Terminal: Activate the Python 3.5 environment, run python, and then import chainer.
* JupyterHub: Connect to JupyterHub, and then go to the Chainer directory to find sample notebooks.

CUDA, cuDNN, NVIDIA Driver

Version(s) supported 10.0.130
Supported DSVM editions Windows and Linux
How is it configured / installed on the DSVM? nvidia-smi is available on the system path.
How to run it Open a command prompt (on Windows) or a terminal (on Linux), and then run nvidia-smi.

Horovod

Version(s) supported 0.16.1
Supported DSVM editions Linux (Ubuntu)
How is it configured / installed on the DSVM? Horovod is installed in Python 3.5
How to run it Activate the correct environment at the terminal, and then run Python.

Keras

Version(s) supported 2.2.4
Supported DSVM editions Windows and Linux
How is it configured / installed on the DSVM? Keras is installed in Python 3.6 on Windows and in Python 3.5 in Linux
How to run it Activate the correct environment at the terminal, and then run Python.

Microsoft Cognitive Toolkit (CNTK)

Version(s) supported 2.5.1
Supported DSVM editions Windows and Linux
How is it configured / installed on the DSVM? CNTK is installed in Python 3.6 on Windows 2016 and in Python 3.5 on Linux)
How to run it Terminal: Activate the correct environment and run Python.
Jupyter: Connect to Jupyter or JupyterHub, and then open the CNTK directory for samples.

MXNet

Version(s) supported 1.3.0
Supported DSVM editions Windows and Linux
How is it configured / installed on the DSVM? MXNet is installed in C:\dsvm\tools\mxnet on Windows and /dsvm/tools/mxnet on Ubuntu. Python bindings are installed in Python 3.6 on Windows 2016 and in Python 3.5 on Linux) R bindings are also included in the Ubuntu DSVM.
How to run it Terminal: Activate the correct conda environment, then run import mxnet.
Jupyter: Connect to Jupyter or JupyterHub, and then open the mxnet directory for samples.

MXNet Model Server

Version(s) supported 1.0.1
Supported DSVM editions Windows and Linux
How is it configured / installed on the DSVM? MXNet Model Server is installed in Python 3.6 on Windows 2016 and in Python 3.5 on Linux)
How to run it Terminal: Run sudo systemctl stop jupyterhub to stop the JupyterHub service first, because both listen on the same port. Then activate the correct conda environment and run mxnet-model-server --start --models squeezenet=https://s3.amazonaws.com/model-server/model_archive_1.0/squeezenet_v1.1.mar

NVidia System Management Interface (nvidia-smi)

Version(s) supported
Supported DSVM editions Windows and Linux
What is it for? NVIDIA tool for querying GPU activity
How is it configured / installed on the DSVM? nvidia-smi is on the system path.
How to run it On a virtual machine with GPU's, open a command prompt (on Windows) or a terminal (on Linux), and then run nvidia-smi.

PyTorch

Version(s) supported 1.2.0
Supported DSVM editions Linux
How is it configured / installed on the DSVM? Installed in Python 3.5. Sample Jupyter notebooks are included, and samples are in /dsvm/samples/pytorch.
How to run it Terminal: Activate the correct environment, and then run Python.
* JupyterHub: Connect, and then open the PyTorch directory for samples.

TensorFlow

Version(s) supported 1.13
Supported DSVM editions Windows, Linux
How is it configured / installed on the DSVM? Installed in Python 3.5 on Linux and Python 3.6 on Windows 2016
How to run it Terminal: Activate the correct environment, and then run Python.
* Jupyter: Connect to Jupyter or JupyterHub, and then open the TensorFlow directory for samples.

TensorFlow Serving

Version(s) supported 1.12
Supported DSVM editions Linux
How is it configured / installed on the DSVM? tensorflow_model_server is available at the terminal.
How to run it Samples are available online.

Theano

Version(s) supported 1.0.3
Supported DSVM editions Linux
How is it configured / installed on the DSVM? Theano is installed in Python 2.7 (root), and in Python 3.5 (py35) environment.
How to run it Terminal: Activate the Python version you want (root or py35), run Python, and then import Theano.
* Jupyter: Select the Python 2.7 or 3.5 kernel, and then import Theano.
To work around a recent math kernel library (MKL) bug, you need to first set the MKL threading layer as follows:

export MKL_THREADING_LAYER=GNU