How long will it take？

[iamwh]

I have tried the model in gpu for about 5 fours but 1 epoch is not finished. So how long will it take to end 1 epoch? Thanks a lot!

[soskek]

Thanks for noticing my code. Unfortunately, the behavior looks natural. This implementation fails to take the advantage of minibatch computation by gpu due to the complexity of its computational graph with many branches. In my experiment on multi-"C"pus, the convergence of training needed a week at least, as shown in a footnote in my paper.