Add the default_weight_observer for the dynamic quantization path (#24231)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/24231
As suggested in https://github.com/pytorch/pytorch/pull/23128#discussion_r309528932, we will add a default weight observer for the dynamic quantization path.
We need to move `observer` and `qconfig` to a separate namespace.
ghstack-source-id: 88583658
Differential Revision: D16781092
fbshipit-source-id: 5cd59c881a7f98b82704ca318b1e63650d73062a
