Environment info

Any

Who can help

@Rocketknight1

Information

BERT TFBertSelfAttention layer calls a function tf.concatenate when given a past_key_value parameter. This function does not exist in the TF 2.0 API. The correct function name is tf.concat.

To reproduce

Call a TFBertLayer layer with a non nil past_key_value argument.

Expected behavior

The code should be modified to use tf.concat instead.