splitting the datasets

Hi, i just wanted to ask if there is any need to split the dataset that I'm gonna use to train, val, and test, or not. since the code has another pipeline to test accuracy using other datasets, will really appreciate any help since I'm really new to this, Thanks!