1.18.23

Fixed

• Correctly create the convolutional embedding layers when the encoder is set to transformer-with-conv-embed. Previously no convolutional layers were added so that a standard Transformer model was trained instead.