diff --git a/paddle/operators/sgd_op.cc b/paddle/operators/sgd_op.cc
index 3bce95535cf10c0df95b503c6e362b3f0ba2e723..8f9eae4186ad848fcecd74b4ab22711f8bb99e2a 100644
--- a/paddle/operators/sgd_op.cc
+++ b/paddle/operators/sgd_op.cc
@@ -27,6 +27,8 @@ class SGDOp : public framework::OperatorWithKernel {
                    "Input(param) of SGDOp should not be null.");
     PADDLE_ENFORCE(ctx->HasInput("grad"),
                    "Input(grad) of SGDOp should not be null.");
+    PADDLE_ENFORCE(ctx->HasInput("learning_rate"),
+                   "Input(learning_rate) of SGDOp should not be null.");
     PADDLE_ENFORCE(ctx->HasOutput("param_out"),
                    "Output(param_out) of SGDOp should not be null.");
 
@@ -42,9 +44,9 @@ class SGDOpMaker : public framework::OpProtoAndCheckerMaker {
   SGDOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
       : OpProtoAndCheckerMaker(proto, op_checker) {
     AddInput("param", "input parameter");
+    AddInput("learning_rate", "learning rate of sgd");
     AddInput("grad", "input gradient");
     AddOutput("param_out", "output parameter");
-    AddAttr<float>("learning_rate", "learning rate of sgd");
     AddComment(R"DOC(
 
 Simplest sgd algorithm.
diff --git a/paddle/operators/sgd_op.h b/paddle/operators/sgd_op.h
index a3fe3308942f98e2c28376b589b6fc930e6878a1..977d201ced31c498c2ab41cf6d412756cabb3aee 100644
--- a/paddle/operators/sgd_op.h
+++ b/paddle/operators/sgd_op.h
@@ -31,7 +31,7 @@ class SGDOpKernel : public framework::OpKernel<T> {
     auto param = ctx.Input<Tensor>("param");
     auto grad = ctx.Input<Tensor>("grad");
     auto param_out = ctx.Output<Tensor>("param_out");
-    float lr = ctx.Attr<float>("learning_rate");
+    float lr = *ctx.Input<float>("learning_rate");
 
     param_out->mutable_data<T>(ctx.GetPlace());
 
diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc
index f4121e9d71824296770f86c1e94c096f767dec0a..d480427f593ccba20ef6555809a524654d0e54b8 100644
--- a/paddle/pybind/pybind.cc
+++ b/paddle/pybind/pybind.cc
@@ -143,6 +143,13 @@ All parameter, weight, gradient are variables in Paddle.
       .def("set_int",
            [](Variable &var, int val) -> void { *var.GetMutable<int>() = val; })
       .def("get_int", [](const Variable &var) -> int { return var.Get<int>(); })
+      .def("is_float", [](const Variable &var) { return var.IsType<float>(); })
+      .def("set_float",
+           [](Variable &var, float val) -> void {
+             *var.GetMutable<float>() = val;
+           })
+      .def("get_float",
+           [](const Variable &var) -> float { return var.Get<float>(); })
       .def("get_tensor",
            [](Variable &self) -> LoDTensor * {
              return self.GetMutable<LoDTensor>();
diff --git a/python/paddle/v2/framework/tests/op_test.py b/python/paddle/v2/framework/tests/op_test.py
index 75df2eeddfe67269d4709887c7cfdb8fab108bd8..81067f38bbf64ac1ab4ccf02aa43b0a38b7d48ad 100644
--- a/python/paddle/v2/framework/tests/op_test.py
+++ b/python/paddle/v2/framework/tests/op_test.py
@@ -46,12 +46,17 @@ def create_op(scope, op_type, inputs, outputs, attrs):
 
 def set_input(scope, op, inputs, place):
     def __set_input__(var_name, var):
-        tensor = scope.find_var(var_name).get_tensor()
-        if isinstance(var, tuple):
-            tensor.set_lod(var[1])
-            var = var[0]
-        tensor.set_dims(var.shape)
-        tensor.set(var, place)
+        if isinstance(var, tuple) or isinstance(var, np.ndarray):
+            tensor = scope.find_var(var_name).get_tensor()
+            if isinstance(var, tuple):
+                tensor.set_lod(var[1])
+                var = var[0]
+            tensor.set_dims(var.shape)
+            tensor.set(var, place)
+        elif isinstance(var, float):
+            scope.find_var(var_name).set_float(var)
+        elif isinstance(var, int):
+            scope.find_var(var_name).set_int(var)
 
     for in_name, in_dup in Operator.get_op_inputs(op.type()):
         if in_name in inputs:
diff --git a/python/paddle/v2/framework/tests/test_sgd_op.py b/python/paddle/v2/framework/tests/test_sgd_op.py
index 64e54d1500c1bc134cc1efe33d41a16dbc08f2d4..f1125f4edb5248abb2a0128a7a8b8b3647ed3317 100644
--- a/python/paddle/v2/framework/tests/test_sgd_op.py
+++ b/python/paddle/v2/framework/tests/test_sgd_op.py
@@ -10,8 +10,7 @@ class TestSGDOp(OpTest):
         g = np.random.random((102, 105)).astype("float32")
         lr = 0.1
 
-        self.inputs = {'param': w, 'grad': g}
-        self.attrs = {'learning_rate': lr}
+        self.inputs = {'param': w, 'grad': g, 'learning_rate': lr}
         self.outputs = {'param_out': w - lr * g}
 
     def test_check_output(self):