diff --git a/paddle/phi/infermeta/spmd_rules/numel.cc b/paddle/phi/infermeta/spmd_rules/numel.cc
index d4a441426296e..76c41ba881ae4 100644
--- a/paddle/phi/infermeta/spmd_rules/numel.cc
+++ b/paddle/phi/infermeta/spmd_rules/numel.cc
@@ -35,7 +35,7 @@ SpmdInfo NumelInferSpmd(const DistMetaTensor& x) {
                         "dims_mapping size [%d] are not matched.",
                         x_ndim,
                         x_dims_mapping.size()));
-  TensorDistAttr out_dist_attr;
+  TensorDistAttr out_dist_attr = CopyTensorDistAttrForOutput(x_dist_attr_src);
   out_dist_attr.set_dims_mapping({});
   std::vector<int64_t> partial_on_dims;
   const auto& dim_mapping = x_dims_mapping;
diff --git a/python/paddle/base/dygraph/tensor_patch_methods.py b/python/paddle/base/dygraph/tensor_patch_methods.py
index d35383c056d2d..5102055c7bdb1 100644
--- a/python/paddle/base/dygraph/tensor_patch_methods.py
+++ b/python/paddle/base/dygraph/tensor_patch_methods.py
@@ -942,6 +942,11 @@ def __nonzero__(self: Tensor) -> bool:
         assert (
             numel == 1
         ), "When Variable is used as the condition of if/while , Variable can only contain one element."
+        # resolve the error issue in scenario of pipeline parallel
+        # where some devices do not have this data, return True or False does not affect
+        # the execution result in those devices, so currently we return False
+        if self.is_dist() and not self._is_initialized():
+            return False
         assert self._is_initialized(), "tensor not initialized"
         return bool(np.array(self) > 0)