The result in conversational task is better than paper's

[C-hase]

I ran your code successfully. Although the result in recommendation part is almost the same, the result of conversational part is so much better than your paper's. And I didn't change the source code. Would you tell me the reasons about it?
The following is the result in my device.
'test/dist@2': 0.5503033723719486, 'test/dist@3': 0.9362212501763792, 'test/dist@4': 1.211090729504727
The following is the result in your paper.

[dayuyang1999]

I am not the author. But here is my guess: Dist is a metric promoting diversity. The sampling method to use and their hyperparameters of generative module can largely influence the diversity. So try different hyperparameters or seed may help.

[wxl1999]

@dayuyang1999 You are right. In addition, Dist is sensitive to the training steps. So I suggest paying more attention to human evaluation.