Reproducing Paper Results

[taehyunzzz]

Hi! Thank you for sharing your implementation.
I have a question on code performance. Were you able to reproduce any of the results provided in the paper?
I have a fastformer code of my own written in pytorch, but I could not reproduce the performance (speed-wise) given in the paper.
So I was wondering if you have tried comparing processing speed with vanilla transformer or any of its variants, for any general task?