Remove cross-rank synchronization during checkpoint load & deprecate torch.distributed.checkpoint.state_dict_loader.load_state_dict#2864
Merged
asolergi-nv merged 7 commits intoNVIDIA:mainfrom Apr 10, 2026
Conversation
Contributor
|
Is |
Contributor
Author
|
They added the warning +2 years ago in pytorch/pytorch#113867.
|
deepakn94
approved these changes
Jan 20, 2026
Contributor
Author
|
/ok to test bd1980c |
Contributor
Author
|
/ok to test dee26a0 |
Contributor
Author
|
/ok to test a37c65b |
Contributor
Author
|
/ok to test a10ca97 |
Contributor
Author
|
/ok to test e835a8c |
ananthsub
approved these changes
Apr 10, 2026
svcnvidia-nemo-ci
added
Approved
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24246552837 |
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24249665766 |
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/24251643090 |
ananthsub
pushed a commit
to ananthsub/Megatron-LM
that referenced
this pull request
Apr 10, 2026
…torch.distributed.checkpoint.state_dict_loader.load_state_dict (NVIDIA#2864)
ananthsub
added a commit
to ananthsub/Megatron-Bridge
that referenced
this pull request
Apr 10, 2026
Cherry-pick of NVIDIA/Megatron-LM#2864 (commit 0602523): - Use checkpoint.load instead of deprecated checkpoint.load_state_dict - Set no_dist=True to remove cross-rank synchronization during checkpoint load Made-with: Cursor
ananthsub
added a commit
to NVIDIA-NeMo/Megatron-Bridge
that referenced
this pull request
Apr 10, 2026
Cherry-pick of NVIDIA/Megatron-LM#2864 (commit 0602523): - Use checkpoint.load instead of deprecated checkpoint.load_state_dict - Set no_dist=True to remove cross-rank synchronization during checkpoint load Made-with: Cursor
ananthsub
added a commit
to ananthsub/Megatron-LM
that referenced
this pull request
Apr 11, 2026
…load The cherry-pick of NVIDIA#2864 incorrectly included the async_strategy parameter in _get_filesystem_reader, which only exists on upstream main but not on ultra-v3-posttraining. Made-with: Cursor
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
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.
no_dist=Trueduring checkpoint load to remove cross-rank synchronization during checkpoint load.torch.distributed.checkpoint.state_dict_loader.loadinstead oftorch.distributed.checkpoint.state_dict_loader.load_state_dict