Ah yes, I missed the part about the size of the dataset. I'm running with a 3GB corpus of text.