Compress all non-Tensor components of a serialized TorchScript model. (#23723)
Summary:
This saves about 69KB off the FaceBlaze model, bringing the total size down from 388KB to 319KB.
See https://github.com/pytorch/pytorch/issues/23582
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23723
Differential Revision: D16623693
fbshipit-source-id: 66267f87635c502c804293054fd5716d291389c0