pytorch/torch/distributed/algorithms
Yi Wang cc4036905c [Gradient Compression] Update the default value of start_powerSGD_iter and update the docstring (#55272)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55272

1. Set 1K as the default value of `start_powerSGD_iter` for practicability. The original default value 10 is usually too small for real use cases. The new default value 1K is also consistent with PyTorch Lightning.
2. Update the docstring of `start_powerSGD_iter` to remind the users to set a value no less than the warm-up steps if any.
3. Update some unit tests to start PowerSGD early.

ghstack-source-id: 125707662

Test Plan: waitforbuildbot

Reviewed By: shuyingsunshine21

Differential Revision: D27553388

fbshipit-source-id: 40076419bc85755c0c0b64b79ba914b241085fcc
2021-04-06 01:27:29 -07:00
..
ddp_comm_hooks [Gradient Compression] Update the default value of start_powerSGD_iter and update the docstring (#55272) 2021-04-06 01:27:29 -07:00
__init__.py [Gradient Compression] Add unit tests that test default Python comm hook implementations (#47158) 2020-11-06 00:28:09 -08:00