CNTK is the open source deep learning toolkit of Microsoft. When learning a model with Brainscript, the application CNTK.EXE produces ".dnn" files which contain the deep learning models, but parameters (aka tensors) and the logic describing the computational network itself. The goal of this job is to produce a C#/.NET library that extracts the tensors from a .dnn file.
Deliverable: a C# library in .NET Core 2.1 that can parse a .dnn file produced by CNTK 2.5 when executed against BrainScript, and that produce a list of named tensors from a .dnn file. A tensor has (1) a name (2) dimensions (3) an array of float values.
We do not expect you to parse all the logic that goes into the model file (e.g. operations), only extract the tensors.