mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
[DDP Note] Remove refs to RoundRobin PG until we officially support it (#40380)
Summary: Removes line mentioning `ProcessGroupRoundRobin` since we don't intend it to be used as a public API just yet. We can add this back when we officially support the API Pull Request resolved: https://github.com/pytorch/pytorch/pull/40380 Differential Revision: D22165556 Pulled By: rohan-varma fbshipit-source-id: 24d0477d881dc74f2ff579de61dfd1ced2b09e75
This commit is contained in:
parent
016cf7d66e
commit
ae2f1f0372
|
|
@ -145,14 +145,11 @@ ProcessGroup
|
|||
|
||||
- `ProcessGroup.hpp <https://github.com/pytorch/pytorch/blob/v1.4.0/torch/lib/c10d/ProcessGroup.hpp>`__:
|
||||
contains the abstract API of all process group implementations. The ``c10d``
|
||||
library provides 4 implementations out of the box, namely,
|
||||
`ProcessGroupGloo`, `ProcessGroupNCCL`, `ProcessGroupMPI`, and
|
||||
`ProcessGroupRoundRobin`, where `ProcessGroupRoundRobin` is a composition of
|
||||
multiple process group instances and launches collective communications in a
|
||||
round-robin manner. ``DistributedDataParallel`` uses
|
||||
``ProcessGroup::broadcast()`` to send model states from the process with rank
|
||||
0 to others during initialization and ``ProcessGroup::allreduce()`` to sum
|
||||
gradients.
|
||||
library provides 3 implementations out of the box, namely,
|
||||
`ProcessGroupGloo`, `ProcessGroupNCCL`, and `ProcessGroupMPI`.
|
||||
``DistributedDataParallel`` uses ``ProcessGroup::broadcast()`` to send
|
||||
model states from the process with rank 0 to others during initialization
|
||||
and ``ProcessGroup::allreduce()`` to sum gradients.
|
||||
|
||||
|
||||
- `Store.hpp <https://github.com/pytorch/pytorch/blob/v1.4.0/torch/lib/c10d/Store.hpp>`__:
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user