Tweeted By @jeremyphoward

Thanks to @apaszke for convincing me to try @PyTorch's jit again - turns out that we get the same speed from memory efficient swish vs the original memory inefficient one now! :) https://t.co/4pMD6kulbr pic.twitter.com/ZBHQEB5xSj

— Jeremy Howard (@jeremyphoward) October 27, 2019