Stanford NLI Corpus (SNLI) [bf337ba8] Tags: nli, datasets 550k examples big enough to use NN's without pre-training caption-based