Question about the predicted reply from trained model

[RookieZB]

Hello. I pretty like your advanced work on this dialogue system. I have tried to train a Mem2Seq model recently, and I get predicted replys shown below, which seems not quite ideal as examples shown on your article. I just want to make sure whether I have made mistakes. I used the recommended hyper parameters from article, and I just printed the outputs when evaluating the models. Thank you.

Here are all predicted replys from first 5 dialogues of test set of KVR:
what city do you want the temperature for
in alhambra temperature on friday low of 20f and high of 30f on friday
the temperature in alhambra will be low and high and of on and on and on and on and on and on and on and on on and on on and on on and on on and on on on and on sunday and
you re welcome
the weather 48 hours in fresno will be a low temperature of of of the tomorrow
you re welcome
there is heavy_traffic on our way but you should be able to reach there in a few minutes
i don t see your friend s home
chevron is moderate_traffic on the route to
chevron is located at 783_arcadia_pl
you re welcome
the closest grocery_store is sigona_farmers_market 4_miles away
sigona_farmers_market is the closest at
you re welcome
what time and time should i to to to to to take your medicine
okay i will to take to you to to take your medicine at 7pm