Batch allreduce ops #220
Merged
Batch allreduce ops #220
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
cswinter
commented
Apr 16, 2022
| @@ -555,9 +555,17 @@ def init_process(xp_info: Any, backend: str = "gloo") -> None: | |||
|
|
|||
|
|
|||
| def gradient_allreduce(model: Any) -> None: | |||
There was a problem hiding this comment.
@vwxyzjn simple change that significantly improves performance
There was a problem hiding this comment.
Oh nice! Good to know. I am gonna give it a try!
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Perform a single allreduce operation over all parameters, which significantly reduces overhead and gives much better performance, especially with many data-parallel replicas. On basic tests I ran, performance matched torch DistributedDataParallel implementation.
Current implementation hits a nice sweet spot of simplicity and performance. There are more opportunities for speedups (smartly grouping parameters and running allreduce in parallel with backward pass), but exploiting these is much more involved and would probably require pulling in horovod or somesuch framework.