Skip to content

[Dev] fix(megatron-fsdp): Resolve hang caused by non-deterministic reduce-scatter #2252

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Nov 20, 2025
Merged

[Dev] fix(megatron-fsdp): Resolve hang caused by non-deterministic reduce-scatter #2252

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
54b10ff
[Megatron-FSDP] Fix hang caused by non-deterministic reduce-scatter
shjwudp Nov 13, 2025
138ee3c
Merge branch 'dev' into fix_mfsdp_random_hang
xuwchen Nov 17, 2025
4f52e0c
Merge branch 'dev' into fix_mfsdp_random_hang
yanring Nov 20, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions megatron/core/distributed/fsdp/src/megatron_fsdp/param_and_grad_buffer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2782,6 +2782,9 @@ def reduce_gradients(
outer_fsdp_group_grad_reduce (bool, optional): Whether to reduce gradients
across outer-DP groups. Defaults to False.
"""
# Sort parameters by their bucket IDs to ensure a deterministic processing order.
# Performing reduce-scatter operations out of order can lead to hangs.
params = sorted(list(params), key=lambda x: self.buffer.param_to_param_group[x])
Copy link
Contributor

@Skylion007 Skylion007 Nov 17, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think the cast to list is necessary, sorted can take an an Iterable including a set.

Suggested change
params = sorted(list(params), key=lambda x: self.buffer.param_to_param_group[x])
params = sorted(params, key=lambda x: self.buffer.param_to_param_group[x])

for param in params:
bucket_id = self.buffer.param_to_param_group[param]
param_group = self.buffer.parameter_groups[bucket_id]
Expand Down
Loading