You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the Transformer, a weight sharing scheme between the input embedding and output projection layer is used to improve efficiency. Any reasons why this is not implemented, and how it could be done?
In the Transformer, a weight sharing scheme between the input embedding and output projection layer is used to improve efficiency. Any reasons why this is not implemented, and how it could be done?