The MontiAnna generator toolchain is capable of generating C++ and Python code based on several well-known frameworks given a NN model architecture, a training configuration, and a dataset. The code generation is delegated by the EMADL Generator. The EMADL Generator forward the composed model to the chosen framework-specific generators, called CNNArch2Gluon, CNNArch2MXNet, CNNArch2Caffe2 and CNNArch2TensorFlow. All the framework-specific generators are an extension of the generator CNNArch2X, which facilitates the extension with further framework generators besides Gluon, MXNet, Caffe2, and TensorFlow. Each of the generators will generate a network creator, network trainer, and network API artifact in Python or C++.
