Ngram based Langauge Models learned using an Encoder-Decoder Model

I have been going through a Ngram based Langauge Model learned using an Encoder-Decoder Model for Email smart compose.

The program output only 1 prediction for given input.
I want to know how to get multiple predictions out of the same.

Here is the link to the notebook: https://nbviewer.jupyter.org/github/PrithivirajDamodaran/NLP-Experiments/blob/master/Gmail_style_smart_compose_with_char_ngram_based_language_model.ipynb

Here, for input sequence : “hi there” the predicted sequence is: “, how are you today?”
But it maybe possible that I have multiple sentences starting with “hi there” in my training dataset. So how do I get all of those?