Loïc

Fix the Bert2Bert call when padded_decode is set to True

1

# Description As mentioned in the following [issue](https://github.com/tensorflow/models/issues/10222), it was impossible to call a Bert2Bert model with the configuration option `padded_decode` set to True. It comes from how the beam...

Add a dedicated padding token to beam search to avoid padding with the start sentence token

1

# Description So far, the beam search module pads its predictions using the initial token ids, it is more modular and easy to use to have the capacity to specify...