Facebook announced that it will open source the PyTorch 1.0 framework in recent months, which is based on the combination of PyTorch and Caffe2, allowing developers to move from research to production without having to migrate.
Last year, Facebook and Microsoft partnered to open up the Open Neural Network Exchange (ONNX) system, which makes it easier to share information between Caffe2 and PyTorch. Now that ONNX has been integrated into PyTorch 1.0, ONNX will be used as a model output format in PyTorch 1.0 so that models can interoperate with other frameworks.
Facebook also stated that it will unify PyTorch 0.4 and Caffe2’s code base in the coming months to create a unified framework that supports multiple functions, including efficient graphics mode execution, analysis, mobile deployment and extensive vendor integration.
PyTorch 1.0 will be released later this year. The company said that Microsoft plans to support the framework in Azure, and Amazon’s network services will support PyTorch 1.0.
Source: TechCrunch
