Due to the time, data, processing power, and memory requirements involved in training sophisticated models, it's often desirable to train models on the server rather than the client. Depending on the use case, even the evaluation of models may need to occur on the server.
There are a few paradigms to consider in terms of where to train and where to evaluate models. Your options, in general, will be to train and evaluate fully on the server, train and evaluate fully on the client, or to train on the server but evaluate on the client. Let's explore some examples of each paradigm.
The simplest implementation is to both train and evaluate models on the server. The main advantage of this approach is that you get ...
