diff --git a/src/operator/tensor/matrix_op-inl.h b/src/operator/tensor/matrix_op-inl.h
index 49847502258f..597c92a8b124 100644
--- a/src/operator/tensor/matrix_op-inl.h
+++ b/src/operator/tensor/matrix_op-inl.h
@@ -1481,7 +1481,7 @@ struct ClipParam : public dmlc::Parameter<ClipParam> {
 struct clip {
   template<typename DType>
   MSHADOW_XINLINE static void Map(index_t i, DType* out, const DType* datas,
-                                  DType a_min, DType a_max) {
+                                  const float a_min, const float a_max) {
     DType data = datas[i];
     if (data > a_max) {
       out[i] = a_max;
@@ -1497,7 +1497,7 @@ struct clip {
 struct clip_grad {
   template<typename DType>
   MSHADOW_XINLINE static void Map(index_t i, DType* out, const DType* grad, const DType* datas,
-                                  DType a_min, DType a_max) {
+                                  const float a_min, const float a_max) {
     DType data = datas[i];
     if (data > a_max) {
       out[i] = 0;
@@ -1524,7 +1524,7 @@ void Clip(const nnvm::NodeAttrs& attrs,
   MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
     mxnet_op::Kernel<mxnet::op::clip, xpu>::Launch(s, outputs[0].Size(), outputs[0].dptr<DType>(),
                                                    inputs[0].dptr<DType>(),
-                                                   DType(param.a_min), DType(param.a_max));
+                                                   param.a_min, param.a_max);
   });
 }
 
@@ -1553,7 +1553,7 @@ void ClipGrad_(const nnvm::NodeAttrs& attrs,
   Stream<xpu> *s = ctx.get_stream<xpu>();
   MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
     Kernel<clip_grad, xpu>::Launch(s, outputs[0].Size(), outputs[0].dptr<DType>(),
-    inputs[0].dptr<DType>(), inputs[1].dptr<DType>(), DType(param.a_min), DType(param.a_max));
+    inputs[0].dptr<DType>(), inputs[1].dptr<DType>(), param.a_min, param.a_max);
   });
 }
 
diff --git a/src/operator/tensor/matrix_op.cc b/src/operator/tensor/matrix_op.cc
index 59e8386d6679..c8e7b60d9548 100644
--- a/src/operator/tensor/matrix_op.cc
+++ b/src/operator/tensor/matrix_op.cc
@@ -702,9 +702,11 @@ MXNET_ADD_SPARSE_OP_ALIAS(clip)
 .describe(R"code(Clips (limits) the values in an array.
 
 Given an interval, values outside the interval are clipped to the interval edges.
-Clipping ``x`` between `a_min` and `a_x` would be::
+Clipping ``x`` between `a_min` and `a_max` would be::
 
-   clip(x, a_min, a_max) = max(min(x, a_max), a_min))
+.. math::
+
+   clip(x, a_min, a_max) = \max(\min(x, a_max), a_min))
 
 Example::
 
@@ -766,7 +768,7 @@ parameter values:
 .add_arguments(ClipParam::__FIELDS__());
 
 NNVM_REGISTER_OP(_backward_clip)
-.set_num_inputs(1)
+.set_num_inputs(2)
 .set_num_outputs(1)
 .set_attr_parser(ParamParser<ClipParam>)
 .set_attr<nnvm::TIsBackward>("TIsBackward", true)
diff --git a/tests/python/unittest/test_operator.py b/tests/python/unittest/test_operator.py
index 157acd252e16..799bf8db99cf 100644
--- a/tests/python/unittest/test_operator.py
+++ b/tests/python/unittest/test_operator.py
@@ -4174,15 +4174,25 @@ def test_special_functions_using_scipy():
 
 
 @with_seed()
-@unittest.skip("Flaky test, tracked at https://github.com/apache/incubator-mxnet/issues/12901")
 def test_clip():
     data = mx.symbol.Variable('data')
     shape = (30, 30)
-    data_tmp = np.random.uniform(-1, 1, shape)
+    data_tmp = np.random.uniform(-1, 1, shape).astype('float32')
     test = mx.sym.clip(data, a_max=0.6, a_min=-0.6)
     check_symbolic_forward(test, [data_tmp], [np.clip(data_tmp, -0.6, 0.6)])
     check_symbolic_backward(test, [data_tmp], [np.ones(shape)],
-                            [np.where(data_tmp < 0.6, [1], [0]) * np.where(data_tmp > -0.6, [1], [0])])
+                            [np.where(data_tmp <= 0.6, [1], [0]) * np.where(data_tmp >= -0.6, [1], [0])])
+
+    # Test monitor on symbol using clip
+
+    def simple_callback(name, arr):
+        pass
+
+    exe = test.simple_bind(ctx=mx.current_context(), data=shape)
+    exe.set_monitor_callback(simple_callback, monitor_all=True)
+    exe.forward(is_train=True)
+    exe.backward(out_grads=mx.nd.ones(shape))
+    mx.nd.waitall()
 
 
 @with_seed()