Figure 1.

The DL Model Training Process

Model training uses a model architecture, weights, and hyperparameters in order to produce model weights. The model architecture and model weights are distributed together as a usable model. Training data and hyperparameters are not, generally, extractable from the model.