4 ms·
in the paper, it seemed to work quite well for LSTMS (language modeling), although it is best suited for tasks with very, very large neural networks for which p
by williamtrask 10y ago
in the paper, it seemed to work quite well for LSTMS (language modeling), although it is best suited for tasks with very, very large neural networks for which parallelism is beneficial