The resulting model.onnx file can then be run on one of the many accelerators that support the ONNX standard.