fix bug (#10862)

Author: zhangbo9674

paddlenlp/transformers/deepseek_v2/modeling_pp.py

@@ -435,7 +435,7 @@ def __init__(self, forward_nodes, backward_nodes, use_fuion=True):
     def forward_backward(self, inputs, output_grad, event_to_wait=None):
         for n in self.nodes:
             inputs, output_grad, event_to_wait = n.forward_backward(inputs, output_grad, event_to_wait)
-        return inputs, output_grad
+        return inputs, output_grad, None
 
 
 class OverlapedScheduleNode: