Extract pruned model for plain inference

[pkhigh]

Hi, amazing work guys. I was wondering if you have some code readily available to extract the fine-tuned pruned sub-model out of the checkpoint model, so that it can be used by simple inference scripts?