Support scripts for PyTorch-ONNX

This repository contains end-to-end tests for PyTorch's ONNX support, including exporting models to Caffe2.

Developing

To run the Caffe2 tests, you'll need a working install of Caffe2. If you're on a devgpu, consider getting a copy via

1. conda install -c ezyang/label/devgpu caffe2 for py3,
2. conda install -c houseroad/label/devgpu caffe2 for py2.7.

# Only needs to be done once
./install-deps.sh
make
make test