pytorch

b35279df - [DDP] Make _ReplicateState inherit from _State and make replicate eagerly initialized (#109647)

[DDP] Make _ReplicateState inherit from _State and make replicate eagerly initialized (#109647) Follow how fully_shard store the _FSDPState, this PR makes _ReplicateState inherit from _State. This PR also makes replicate eagerly initialize the internal DDP instance so that users can access the required methods/functions before the first forward(). Differential Revision: [D49428291](https://our.internmc.facebook.com/intern/diff/D49428291/) Pull Request resolved: https://github.com/pytorch/pytorch/pull/109647 Approved by: https://github.com/wz337, https://github.com/rohan-varma ghstack dependencies: #110688