PROFET is a model that can predict performance without directly revealing the implementation of the CNN internal model architecture. It uses CNN learning profiling information as its predictive model dataset, and uses new heuristic methods for feature engineering and modeling. The prediction accuracy of PROFETs across different GPU instances, batch sizes and input image pixel sizes is practical and well suited for public cloud computing environments with different service providers and consumers.

You can try running PROFET directly through the demo system on this web page. If you look at the documentation part, you can check the architecture of PROFET and learn how to use it, and it provides Docker as an open source. You can communicate with the researcher through the information in Contact.