Add sample

[jxcodetw]

Can you show us some results that sampled from your implementation of the model ?

[paultsw]

Hi jxcodetw,

I'm working on it! I've had convergence issues training with low batch sizes (< 1024) so I'm doing some debugging on this matter to figure out why a batch size of 256 works for the authors and not for me. Once I do that, I'll merge the fixed updates.

This is a side-project for me and not my main day-to-day work, so I can't guarantee how quick it'll be before I can do that (hopefully in the next few days).