How is the training data accuracy calculated? Are we “hiding” the labels of the training data when training our model?

Question

i.e. - when we show our model what the true label is during training, it adjusts hyper parameters in order to perform better and it stops when the loss function hits a min?