about M1,M2,M3 in Context Enhanced Path (CEP)

[xubin1994]

Hello, I read your code,It seems that the code does not reflect three different design choices for Context Enhanced Path (CEP).all the feat use one Transformer with same architecture of self attention and cross attention.Can you explain in detail where M1,M2,M3 is reflected in the code?

[guoweiyu]

Thanks for your question! Now, there is only M3 in open-source code because M3 performs best. If you want to try M1 or M2, you can change the code at line 66 of transformer.py. And we will add M1 and M2 in the future version.