This repository was archived by the owner on Jul 1, 2024. It is now read-only.
Support sync batch norm group size #534
Closed
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
Summary: Sync BatchNorm is an expensive operation when we try to synchronize across multiple nodes. Adding the option of specifying the group size for the sync op. A value of 8 would mean that the sync only happens intra node (since the ranks are aligned in multiples of 8) for an 8 GPU per node setup. This only works with Apex Sync BN. When trying it with PyTorch with a group size of 8, I was getting connection reset errors inside `init_distributed_data_parallel_model`. When the group size equalled the total number of GPUs, I didn't get issues. I think we have too many options at the config level for distributed settings. Moving them to a separate section in a follow up diff. Differential Revision: D21868629 fbshipit-source-id: 0983ae702329c716618407a332e67c00838a3b17
facebook-github-bot
added
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D21868629 |
facebook-github-bot
pushed a commit
that referenced
this pull request
Jun 4, 2020
Summary: Pull Request resolved: #534 Sync BatchNorm is an expensive operation when we try to synchronize across multiple nodes. Adding the option of specifying the group size for the sync op. A value of 8 would mean that the sync only happens intra node (since the ranks are aligned in multiples of 8) for an 8 GPU per node setup. This only works with Apex Sync BN. When trying it with PyTorch with a group size of 8, I was getting connection reset errors inside `init_distributed_data_parallel_model`. When the group size equalled the total number of GPUs, I didn't get issues. I think we have too many options at the config level for distributed settings. Moving them to a separate section in a follow up diff. Reviewed By: vreis Differential Revision: D21868629 fbshipit-source-id: e659d8339a2d03a6cb4be2eb9599223583191c2b
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Summary:
Sync BatchNorm is an expensive operation when we try to synchronize across multiple nodes. Adding the option of specifying the group size for the sync op. A value of 8 would mean that the sync only happens intra node (since the ranks are aligned in multiples of 8) for an 8 GPU per node setup.
This only works with Apex Sync BN. When trying it with PyTorch with a group size of 8, I was getting connection reset errors inside
init_distributed_data_parallel_model. When the group size equalled the total number of GPUs, I didn't get issues.I think we have too many options at the config level for distributed settings. Moving them to a separate section in a follow up diff.
Differential Revision: D21868629