Hi !
The only discussion about pruning I can think of is this one. I don’t see any reason why pruning couldn’t be added to fastai if it is operational.
I am also interested in model pruning/compression. I am currently making experiments where I combine the technique you describe here (which I prefer calling sparsifying) with kernel-level pruning to get benefits in terms of parameters/speed. Currently this works well, I hope we’ll see developments in that way.