Skip to content

Average sparse gradients by world size regardless of gradient_predivide_factor #8772

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

Draft
vineethsaivs wants to merge 1 commit into
base: master
Choose a base branch
from
Draft

Average sparse gradients by world size regardless of gradient_predivide_factor #8772

Show file tree
Hide file tree
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
6 changes: 2 additions & 4 deletions deepspeed/runtime/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3932,10 +3932,8 @@ def sparse_allreduce(self, sparse, dp_group, dp_world_size=None):
if dp_world_size is None:
dp_world_size = dist.get_world_size(group=dp_group)
if self.gradient_average:
if self.postscale_gradients():
values.mul_(self.gradient_predivide_factor() / (dp_world_size))
else:
values.mul_(1. / (dp_world_size))
# all_gather concatenates, so no predivide happened; the full average is 1 / dp_world_size
values.mul_(1. / dp_world_size)

indices_device_list = self.sparse_all_gather(indices, dp_group)
values_device_list = self.sparse_all_gather(values, dp_group)
Expand Down
9 changes: 7 additions & 2 deletions tests/unit/runtime/sparse_tensor/test_averaging_sparse_gradients.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,10 +59,15 @@ def get_data(device):
class TestSparseAdam(DistributedTest):
world_size = 2

def test(self):
@pytest.mark.parametrize("gradient_predivide_factor", [1.0, 2.0])
def test(self, gradient_predivide_factor):
skip_on_arch(min_arch=7)

config_dict = {"train_batch_size": 2, "steps_per_print": 1}
config_dict = {
"train_batch_size": 2,
"steps_per_print": 1,
"gradient_predivide_factor": gradient_predivide_factor
}
model, optimizer = get_model_optimizer()
loss = torch.nn.BCEWithLogitsLoss()
engine, _, _, _ = deepspeed.initialize(model=model, optimizer=optimizer, config=config_dict)
Expand Down