Skip to content

Fix paddle.distributed.broadcast/scatter from async to sync #22

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
takeshi-yoshimura merged 1 commit into from
Jun 9, 2025
Merged

Fix paddle.distributed.broadcast/scatter from async to sync #22

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
8 changes: 4 additions & 4 deletions fastsafetensors/tensor_factory.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ def shuffle(self, pg: dist.ProcessGroup, tensor_name: str, dim: int, group = Non
if self.metadata.framework == "pytorch":
dist.broadcast(dst, self.rank, group=pg)
elif paddle_loaded and self.metadata.framework == "paddle":
pdist.broadcast(dst, self.rank, group=group, sync_op=False)
pdist.broadcast(dst, self.rank, group=group)
else:
rank_slices: List[Tuple] = [() for i in range(0, pg.size())]
size = frame.shape[dim]
Expand Down Expand Up @@ -135,7 +135,7 @@ def shuffle(self, pg: dist.ProcessGroup, tensor_name: str, dim: int, group = Non
if self.metadata.framework == "pytorch":
dist.scatter(dst, scatter_list=scatter_list, src=self.rank, group=pg)
elif paddle_loaded and self.metadata.framework == "paddle":
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group, sync_op=False)
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group)
self.shuffled[tensor_name] = dst
return dst

Expand Down Expand Up @@ -173,7 +173,7 @@ def shuffle_packed_qkv(self, pg: dist.ProcessGroup, tensor_name: str, group = No
dist.scatter(dst, scatter_list=scatter_list, src=self.rank, group=pg)
elif paddle_loaded and self.metadata.framework == "paddle":
dst = paddle.to_tensor(paddle.empty(shape=new_shape, dtype=frame.dtype),place=self.device)
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group, sync_op=False)
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group)
self.shuffled[tensor_name] = dst
return dst

Expand Down Expand Up @@ -214,7 +214,7 @@ def shuffle_multi_cols(self, pg: dist.ProcessGroup, tensor_names: List[str], dim
dist.scatter(dst, scatter_list=scatter_list, src=self.rank, group=pg)
elif paddle_loaded and self.metadata.framework == "paddle":
dst = paddle.to_tensor(paddle.empty(shape=new_shape, dtype=frame.dtype), place=self.device )# dst should be eariler than scatter_list for less fragmentation
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group, sync_op=False)
pdist.scatter(dst, tensor_list=scatter_list, src=self.rank, group=group)
return dst

def free_dev_ptrs(self):
Expand Down
Loading