Why not SNLI?

#6
by umarbutler - opened

Hey @MoritzLaurer , thanks for creating this wonderful model. I'm just wondering why you didn't include SNLI in your training data. Did you find that it was too noisy, not diverse enough or was there some other reason?

yes, exactly, SNLI turned out to be too noisy and texts are all very short with a limited domain coverage. My general lesson from training these models is: quality > quantity.

Sign up or log in to comment