Fix the bug of sd3 controlnet training when using gradient checkpointing. (#9498)

Fix the bug of sd3 controlnet training when using gradient_checkpointing. Refer to issue #9496

src/diffusers/models/controlnet_sd3.py

@@ -336,7 +336,7 @@ class SD3ControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginal
                     return custom_forward
 
                 ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
-                hidden_states = torch.utils.checkpoint.checkpoint(
+                encoder_hidden_states, hidden_states = torch.utils.checkpoint.checkpoint(
                     create_custom_forward(block),
                     hidden_states,
                     encoder_hidden_states,