Comment by niemandhier

Currently we do not explore sparsity. The next iteration of models will be much more compact by focusing on reducing effective tensor size.