overfitting about crcn

[cesc-park]

3 stages mean 15 epochs, that's enough I think.
It would be nice to test it by reducing the size of the model.

[YangXS]

Thank you for your reply.
I tested the model trained with 3 stages.
The evaluated results are still not good:
Bleu_1: 0.127
Bleu_2: 0.032
Bleu_3: 0.007
Bleu_4: 0.002

Could you give more details about how to reduce the size of the model ?

[cesc-park]

Did you set the same hyper-parameter settings as in the paper?

[cesc-park]

The code is for demonstrating examples. To make it work faster, there are smaller hyper-parameter values than the numbers in the paper. ex) TOPK = 3

[YangXS]

I actually only used the default hyper-parameter setting in the training and generation code.
I will try to fix that.
Thanks.