you're right, I should say just the model, or trained model. what terminology would you use? For example, with stable diffusion I download a.... model? and can also train a model?
You might tangentially download the code to train a new model from scratch or update the model you received.
The important bit is that the training data is HUGE and the model is not.