loss computation and labels preparation

[kunalpagarey]

Hi @pengbaolin, in paper it is mentioned that loss is computed only for the target text (x'). But in your code I can't understand why the inputs and labels are exactly same (Dialogue act + target text) for training? I guess the Dialogue act part should be removed or target text should be masked to create the labels.
Is there anything I am missing out in the code?

Please help me to understand this.

Thank you.

[lytum]

Hi @pengbaolin, in paper it is mentioned that loss is computed only for the target text (x'). But in your code I can't understand why the inputs and labels are exactly same (Dialogue act + target text) for training? I guess the Dialogue act part should be removed or target text should be masked to create the labels.
Is there anything I am missing out in the code?

Please help me to understand this.

Thank you.

Hi, I also have the same question, may I ask you, have you already understood? Thanks for any information.