Skip to content

fix: skip FSDP DTensor boundary validation under fake process group #3686

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

Open
Victarry wants to merge 1 commit into
base: main
Choose a base branch
from
+5 −0
Open

fix: skip FSDP DTensor boundary validation under fake process group #3686

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 5 additions & 0 deletions megatron/core/distributed/fsdp/src/megatron_fsdp/uneven_dtensor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -175,6 +175,11 @@ def validate_uneven_dtensor(dtensor: DTensor) -> None:
)

# Check that all boundaries (start and end) are touched.
# Skip under fake process group — all_reduce is a no-op so only rank 0's
# boundaries are visible, which makes the end-boundary check always fail.
if torch.distributed.is_initialized() and torch.distributed.get_backend() == 'fake':
return

boundary_checks = torch.tensor(
[
[offset == 0, offset + size == dtensor.shape[dim]]
Expand Down
Loading