Small typos in functions set_none_gradients_to_zero (#5557)

change from "zero_like" to "zeros_like"
microsoft · May 21, 2024 · 975290a · 975290a
1 parent 5b314f4
commit 975290a
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/deepspeed/runtime/zero/stage3.py b/deepspeed/runtime/zero/stage3.py
@@ -1575,7 +1575,7 @@ def set_none_gradients_to_zero(self, i, partition_id):
         for param_id in self.is_grad_computed[i][partition_id]:
             param = self.param_dict[param_id]
             if param.grad is None:
-                param.grad = torch.zero_like(param)
+                param.grad = torch.zeros_like(param)
 
     ######################Reduction Related Methods##############################
 

diff --git a/deepspeed/runtime/zero/stage_1_and_2.py b/deepspeed/runtime/zero/stage_1_and_2.py
@@ -1474,7 +1474,7 @@ def set_none_gradients_to_zero(self, i, partition_id):
         for param_id in self.is_grad_computed[i][partition_id]:
             param = self.param_dict[param_id]
             if param.grad is None:
-                param.grad = torch.zero_like(param)
+                param.grad = torch.zeros_like(param)
 
     ######################Reduction Related Methods##############################
     def allreduce_bucket(self, bucket, rank=None, log=None, divide=True, process_group=None):