diff --git a/paddle/fluid/pybind/op_function_generator.h b/paddle/fluid/pybind/op_function_generator.h
index d9aab3dbb04ce4a129ef2d5407ec18a8219c42a2..c348e04e6c7acbecdd36099bc9dbeceba8262229 100644
--- a/paddle/fluid/pybind/op_function_generator.h
+++ b/paddle/fluid/pybind/op_function_generator.h
@@ -227,7 +227,6 @@ std::map<std::string, std::set<std::string>> op_passing_outs_map = {
     {"c_reduce", {"Out"}},
     {"c_scatter", {"Out"}},
     {"barrier", {"Out"}},
-    {"assign", {"Out"}},
     {"fake_quantize_dequantize_moving_average_abs_max",
      {"Out", "OutScale", "OutAccum", "OutState"}},
     {"fake_quantize_dequantize_abs_max", {"Out", "OutScale"}},
@@ -243,6 +242,8 @@ std::map<std::string, std::set<std::string>> op_passing_outs_map = {
     {"get_float_status", {"FloatStatusOut"}},
     {"assign", {"Out"}},
     {"assign_value", {"Out"}},
+    {"split", {"Out"}},
+    {"concat", {"Out"}},
 };
 
 // NOTE(pangyoki): Tensor View Strategy.
diff --git a/python/paddle/fluid/layers/nn.py b/python/paddle/fluid/layers/nn.py
index 7dc0d0af68c16359854915236a1d2aed71c34c77..9be15d23bb371a977c3f78cd178dbe5bd2cccc24 100755
--- a/python/paddle/fluid/layers/nn.py
+++ b/python/paddle/fluid/layers/nn.py
@@ -5026,7 +5026,9 @@ def split(input, num_or_sections, dim=-1, name=None):
             raise TypeError(
                 "The type of 'num_or_sections' in split must be int, list or tuple in imperative mode, but "
                 "received %s." % (type(num_or_sections)))
-        return _C_ops.split(input, num, *attrs)
+        out = [_varbase_creator() for n in range(num)]
+        _C_ops.split(input, out, *attrs)
+        return out
 
     check_variable_and_dtype(
         input, 'input',
diff --git a/python/paddle/fluid/layers/tensor.py b/python/paddle/fluid/layers/tensor.py
index a49b4b79fbf0ccef4d5075dd286fbd92c1891388..1cac55170476fd85e45932788ff9db288faf4b84 100644
--- a/python/paddle/fluid/layers/tensor.py
+++ b/python/paddle/fluid/layers/tensor.py
@@ -337,7 +337,9 @@ def concat(input, axis=0, name=None):
             axis = axis.item(0)
         if not isinstance(input, Variable):
             input = [t for t in input if t.shape.count(0) == 0]
-        return _C_ops.concat(input, 'axis', axis)
+        out = _varbase_creator()
+        _C_ops.concat(input, out, 'axis', axis)
+        return out
 
     check_type(input, 'input', (list, tuple, Variable), 'concat')
     if not isinstance(input, Variable):
diff --git a/python/paddle/nn/utils/transform_parameters.py b/python/paddle/nn/utils/transform_parameters.py
index ef5cd8700761f8c5c2ab73c2ec4c88d5475f831d..99870ce29a138dab7bd52e76fd1e582a8ffc045e 100644
--- a/python/paddle/nn/utils/transform_parameters.py
+++ b/python/paddle/nn/utils/transform_parameters.py
@@ -69,7 +69,9 @@ def parameters_to_vector(parameters, name=None):
     out = _varbase_creator(dtype=dtype)
     if in_dygraph_mode():
         with paddle.fluid.dygraph.no_grad():
-            _C_ops.concat(parameters, 'axis', 0)._share_underline_tensor_to(out)
+            tmp = _varbase_creator()
+            _C_ops.concat(parameters, tmp, 'axis', 0)
+            tmp._share_underline_tensor_to(out)
     else:
         _dygraph_tracer().trace_op(
             type='concat',
@@ -120,8 +122,8 @@ def vector_to_parameters(vec, parameters, name=None):
 
     if in_dygraph_mode():
         with paddle.fluid.dygraph.no_grad():
-            res = _C_ops.split(vec,
-                               len(parameters), 'axis', 0, 'sections', sections)
+            res = [_varbase_creator() for n in range(len(parameters))]
+            _C_ops.split(vec, res, 'axis', 0, 'sections', sections)
             for i in range(0, len(res)):
                 res[i]._share_underline_tensor_to(parameters[i])
     else:
diff --git a/python/paddle/tensor/math.py b/python/paddle/tensor/math.py
index 5376d393ea4325ed8ce70bd439fd5dd0e0b85caf..c6cbb533c4453e6495ead305877eeafc0fc15f4c 100755
--- a/python/paddle/tensor/math.py
+++ b/python/paddle/tensor/math.py
@@ -3911,7 +3911,8 @@ def diff(x, n=1, axis=-1, prepend=None, append=None, name=None):
             input_list = [x, append]
             has_pend = True
         if has_pend:
-            new_input = _C_ops.concat(input_list, 'axis', axis)
+            new_input = _varbase_creator()
+            _C_ops.concat(input_list, new_input, 'axis', axis)
         else:
             new_input = x