dingo_train and dingo_train_condor seem to use roughly double the necessary amount of CPU RAM at the start of training. This means that we must request much more memory than is ultimately necessary for training.
I suspect this occurs when we call split_dataset_into_train_and_test():
dingo_train
anddingo_train_condor
seem to use roughly double the necessary amount of CPU RAM at the start of training. This means that we must request much more memory than is ultimately necessary for training.I suspect this occurs when we call
split_dataset_into_train_and_test()
:https://github.com/dingo-gw/dingo/blob/bf2c41f7630860268f4e4aed8c0a6fe9d30a13a4/dingo/core/utils/torchutils.py#L232