How does the trainer script portion the dataset for eval/validity check?

[m000lie]

I don't see anywhere in the trainer.py script that mentions fractioning the trainval_dataset.npy for eval, can somebody show me or tell me how much of the dataset is used for eval and how i can change it?