Yeah, it's probably less of a concern for text tasks, where the data per example is relatively light (though there is a whole internet worth of text data...) I

Yeah, it's probably less of a concern for text tasks, where the data per example is relatively light (though there is a whole internet worth of text data...) I mostly work with audio, where individual examples are ~2MB, so the dataset sizes get very heavy quickly.