TransformerEncoderLayer In Embellish Stage Confusion

[AndyWeasley2004]

I have read your paper and the embellish stage should use a transformer decoder-based model, right? However, when I read the code, the decoder class actually is constructed from TransformerEncoderLayer in the fast-transformer package, and in the implementation of this package, the encoder layer is still the encoder, not the decoder, so I'm a little confused about this point. Could you kindly explain how it works?