The Reformer – Pushing the limits of language modeling 6 years agoIMO the theoretical insight w.r.t. transformers as RNNs through the kernel formulation of self-attention is more interesting than the experimental results. 0ThreadHN