In September we introduced the Open Neural Network Exchange (ONNX) format Microsoft co-created to increase interoperability and reduce friction for developing and deploying AI. Through the fall, a number of companies that share our vision announced their support for the ONNX ecosystem and ONNX v1 was released in December.
Today we are excited to share Preferred Networks (PFN) is adding ONNX support in Chainer. ONNX is strategically important to PFN because it can be the solution to bridge Chainer’s powerful, flexible, intuitive characteristics originated from define-by-run approach and ease of deployment to various devices. PFN also recognizes the benefits of the ONNX open ecosystem to enable developers working on deep learning to move between tools easily, choosing ones that are best suited for the task at hand. It’s great to have more framework support for ONNX — Caffe2, PyTorch, Microsoft Cognitive Toolkit, MXNet and now Chainer.
At Microsoft we believe bringing AI advances to all developers, on any platform, using any language, with an open AI ecosystem, will help ensure AI is more accessible and valuable to all. In May, we announced a global alliance partnership with Preferred Networks and jointly drove Chainer and Microsoft AI services integration such as Azure DSVM Chainer preinstall, Chainer x100 performance on 128 Azure InfiniBand GPU Cluster/Azure Resource Manager template for friction free deployment, Chainer and Azure Batch AI integration. With ONNX support for Chainer, developers, data scientists and researchers will be able to deliver new and exciting AI innovations faster.
Link to Preferred Networks Blog