Skip to content
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

Allow scalar all gather #5797

Merged
merged 6 commits into from
Dec 11, 2023
Merged

Allow scalar all gather #5797

Changes from 3 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
129724b
fix scalar all gather
bfontain Nov 2, 2023
346db85
update code
bfontain Nov 6, 2023
54af204
reformat
bfontain Nov 7, 2023
2b054e3
add unit test
bfontain Nov 15, 2023
f02dd44
Merge branch 'master' into fix-scalar-allgather
bfontain Dec 7, 2023
3062f75
fix lint
bfontain Dec 7, 2023
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
6 changes: 5 additions & 1 deletion torch_xla/distributed/xla_backend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,10 +73,14 @@ def allreduce(self, tensors, all_reduce_options):

def allgather(self, output_tensors_list, input_tensors, opts=None):
for input_tensor, output_tensors in zip(input_tensors, output_tensors_list):
is_scalar = (input_tensor.dim() == 0)
if is_scalar:
input_tensor = torch.reshape(input_tensor, (1,))
result = xm.all_gather(input_tensor, groups=self._mesh, pin_layout=False)
for i, slice in enumerate(torch.split(result, input_tensor.shape[0])):
with torch.no_grad():
output_tensors[i].copy_(slice)
output_tensors[i].copy_(
slice if not is_scalar else torch.reshape(slice, ()))

JackCaoG marked this conversation as resolved.
Show resolved Hide resolved
return _ret_work([t for sublist in output_tensors_list for t in sublist])

Expand Down
Loading