Is there a straightforward way to implement relative positional encoding as described in the Shaw paper using Tensorflow instead of absolute positional encoding? Thanks!
Asked
Active
Viewed 33 times
