subwords vocabulary is needed to speed up the repo.

[fiyen]

This is a very useful task and many works are implemented based on it! However, when I use this repo to compare the performance of my work, I find that the generation process is too slow (I'd like to generate thousands of new samples). I checked the vocab and found that the vocab is based on characters. This makes the vocab compatible with other works, however, iteratively generating characters looks a bit costly for producing long sentences. So whether there is a plan to add a subwords vocabulary?