diff --git a/src/diffusers/models/resnet.py b/src/diffusers/models/resnet.py
index c0b4ad40055a..a98cb491146b 100644
--- a/src/diffusers/models/resnet.py
+++ b/src/diffusers/models/resnet.py
@@ -366,7 +366,12 @@ def forward(self, input_tensor: torch.Tensor, temb: torch.Tensor, *args, **kwarg
         hidden_states = self.conv2(hidden_states)
 
         if self.conv_shortcut is not None:
-            input_tensor = self.conv_shortcut(input_tensor.contiguous())
+            # Only use contiguous() during training to avoid DDP gradient stride mismatch warning.
+            # In inference mode (eval or no_grad), skip contiguous() for better performance, especially on CPU.
+            # Issue: https://github.com/huggingface/diffusers/issues/12975
+            if self.training:
+                input_tensor = input_tensor.contiguous()
+            input_tensor = self.conv_shortcut(input_tensor)
 
         output_tensor = (input_tensor + hidden_states) / self.output_scale_factor
 
diff --git a/tests/pipelines/test_pipelines_common.py b/tests/pipelines/test_pipelines_common.py
index 7db5f4da89ca..7d898919dfee 100644
--- a/tests/pipelines/test_pipelines_common.py
+++ b/tests/pipelines/test_pipelines_common.py
@@ -2212,7 +2212,7 @@ def test_StableDiffusionMixin_component(self):
 
     @require_hf_hub_version_greater("0.26.5")
     @require_transformers_version_greater("4.47.1")
-    def test_save_load_dduf(self, atol=1e-4, rtol=1e-4):
+    def test_save_load_dduf(self, atol=1e-3, rtol=1e-3):
         if not self.supports_dduf:
             return