difference between loss computation in sample_validation_data and training

[podragonwer]

when training, whole sequence is put into the net and loss is calculated at once.
in sample_validation_data, loss is accumulated every step.
can you explain why you did this and what difference it makes?
thank you.