Currently, the embedding for the encoder is inside the encoder while the embedding for the decoder is inside the sequence to sequence model. The embedding being present in the decoder would make it easier to turn it on/off and this is particularly relevant given tree to seq wants embeddings in the decoder while tree to seq anc does not want decoder embeddings.
Currently, the embedding for the encoder is inside the encoder while the embedding for the decoder is inside the sequence to sequence model. The embedding being present in the decoder would make it easier to turn it on/off and this is particularly relevant given tree to seq wants embeddings in the decoder while tree to seq anc does not want decoder embeddings.