Skip to content

Fix All-Gather under world size one #7219

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 4 commits into from
Jun 20, 2025
Merged

Fix All-Gather under world size one #7219

Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
6996769
fix gather under world size one
ch-wan Jun 16, 2025
8156980
minor
ch-wan Jun 16, 2025
ddcd88e
add warning info
ch-wan Jun 17, 2025
4ae9d20
Merge branch 'main' into fix/gather-under-world-size-one
zhyncs Jun 17, 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
12 changes: 8 additions & 4 deletions python/sglang/srt/distributed/parallel_state.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -523,17 +523,21 @@ def all_gather(
self,
input_: torch.Tensor,
dim: int = -1,
tensor_list: List[torch.Tensor] = None,
output_tensor_list: Optional[List[torch.Tensor]] = None,
) -> torch.Tensor:
world_size = self.world_size
# Bypass the function if we are using only 1 GPU.
if world_size == 1:
return input_
if output_tensor_list is not None:
output_tensor_list[0].copy_(input_)
return None
else:
return input_

if tensor_list is not None:
if output_tensor_list is not None:
# TODO(ch-wan): support other backends
return torch.distributed.all_gather(
tensor_list, input_, group=self.device_group
output_tensor_list, input_, group=self.device_group
)

assert (
Expand Down
Loading