Running out of memory processing dataset

[jbmaxwell]

I see that there’s a “streaming” option to load_dataset. Would it makes sense to make that True? And will that prevent shuffling the data?