Significance of hyperparameters in defaults.py

[TanayKarve]

I understand that the model uses default hyperparameters for training from defaults.py. However why are these specific values used?
The reason I ask this is, the harvard paper for this model mentions different values for certain parameters, e.g. token embedding size is 80 whereas defaults.py sets it to be 10, then hidden units of attention decoder in the paper is 512, whereas defaults.py sets it as 128.
Are the values in defaults.py optimum for im2latex problem or should i stick with the ones in the paper?

[saymyname77]

just for fun..