To install Nervana Graph, you must first install our neon in a virtual environment. For neon install instructions, see:

Activate the neon virtualenv with . .venv/bin/activate and then run:

git clone
cd ngraph
make install

Getting Started

Several jupyter notebook walk-throughs demonstrate how to use Nervana Graph:

  • ngraph/examples/walk_through/ guides developers through implementing logistic regression with ngraph
  • ngraph/examples/mnist/MNIST_Direct.ipynb demonstrates building a deep learning model using ngraph directly.

The neon frontend can also be used to define and train deep learning models:

  • ngraph/examples/mnist/ Multi-layer perceptron network on MNIST dataset.
  • ngraph/examples/cifar10/ Convolutional neural network on CIFAR-10.
  • ngraph/examples/cifar10/ Multi-layer perceptron on CIFAR-10 dataset.
  • ngraph/examples/ptb/ Character-level RNN model on Penn Treebank data.

We also include examples for using tensorflow to define graphs that are then passed to ngraph for execution:

  • ngraph/frontends/tensorflow/examples/
  • ngraph/frontends/tensorflow/examples/
  • ngraph/frontends/tensorflow/examples/

Developer Guidelines

Before checking in code, run the unit tests and check for style errors:

make test
make style

Documentation can be generated via:

make doc

And viewed at doc/build/html/index.html.