Any idea to delete dataset_Train after "fit"?

[allanchan339]

Due to IO performance, I usually load all image-label pairs in RAM to accelerate the model training.
It is done by creating a large NumPy array in the dataset.init(). This method consumes almost all the RAM. Therefore, I need to manually delete the dataset_Train after training to free some space for dataset_Test.

The GC and del methods work well in PyTorch. However, PyTorch lightning recommends using DataModule to wrap the dataset_Train, dataset_Test and etc. I don't know how to implement the same deletion after "fit".

Any idea to implement the same implementation when using the datamodule to wrap?