Efficient way of loading data from an h5 dataset?

[malfonsoarquimea]

Hi! I am writing the code of a datamodule whose get_item() method will simply perform some operations to chose an index and then picking the item at this index from the h5 dataset.
The thing is that, as far as I know, it will be much slower to pick a batch one by one from the h5 file than picking the whole batch at once. Is there some more clever approach to do this so that the datamodule itself is more efficient (accounting for the dataset and the dataloaders)

Thanks in advance!