Thanks for your fancy code! Your code is helpful to my recent project, which needs two source inputs(X1 and X2). I need the transformer to attend on two sources, while i only want to copy from source X1. What's your opinion about this?
A possible solution is that i use two encoders and one decoder. I only feed the encoder outputs of X1 into the last layer of the decoder, otherwise i feed the concatenation outputs of two encoders for joint attention. Do you think it can work well ? Anything wrong with the solution?
Hi, @policeme
Thanks for your fancy code! Your code is helpful to my recent project, which needs two source inputs(X1 and X2). I need the transformer to attend on two sources, while i only want to copy from source X1. What's your opinion about this?
A possible solution is that i use two encoders and one decoder. I only feed the encoder outputs of X1 into the last layer of the decoder, otherwise i feed the concatenation outputs of two encoders for joint attention. Do you think it can work well ? Anything wrong with the solution?
Looking forward to your insightful opinion!