diff --git a/AUTHORS.md b/AUTHORS.md
index e5481d83de190665d0f835b1180d1c603f4f6de7..a8ea5c46e94d25e276d0a84e731c5b4e09de8aad 100644
--- a/AUTHORS.md
+++ b/AUTHORS.md
@@ -57,6 +57,7 @@
 | reyoung | Yang Yu |
 | [Sand3r-](https://raw.githubusercontent.com/jczaja/Paddle/paddle-poland-team/doc/images/paddle_poland_team.jpg)| Michal Gallus |
 | [sfraczek](https://raw.githubusercontent.com/jakpiase/Paddle/new_paddle_intel_authors/img/img.jpg)| Sylwester Fraczek |
+| Silv3S | Slawomir Siwek |
 | sneaxiy | Jin-Le Zeng |
 | Superjom | Chun-Wei Yan |
 | tensor-tang | Jian Tang |
diff --git a/paddle/fluid/operators/mkldnn/layer_norm_mkldnn_op.cc b/paddle/fluid/operators/mkldnn/layer_norm_mkldnn_op.cc
index 8f98a0b9fbee8666d0959c0eda962a9e864af19f..5b499b8985f4f6c823e6ce41e517bdeecc3b60b7 100644
--- a/paddle/fluid/operators/mkldnn/layer_norm_mkldnn_op.cc
+++ b/paddle/fluid/operators/mkldnn/layer_norm_mkldnn_op.cc
@@ -29,18 +29,10 @@ class LayerNormMKLDNNHandler : public platform::MKLDNNHandlerNoCachingT<
                          const dnnl::engine engine, platform::Place cpu_place)
       : platform::MKLDNNHandlerNoCachingT<T, dnnl::layer_normalization_forward>(
             engine, cpu_place) {
-    if (!is_test) {
-      // TODO(grygielski) Delete forcing stats_md after DNNL 1.2 is introduced
-      auto stats_md = dnnl::memory::desc(
-          {begin(dims), end(dims) - 1}, platform::MKLDNNGetDataType<float>(),
-          platform::GetPlainMKLDNNFormat(dims.size() - 1));
-      this->AcquireForwardPrimitiveDescriptor(dnnl::prop_kind::forward_training,
-                                              x->mem_desc(), stats_md, epsilon,
-                                              flags);
-    } else {
-      this->AcquireForwardPrimitiveDescriptor(
-          dnnl::prop_kind::forward_inference, x->mem_desc(), epsilon, flags);
-    }
+    const auto fwd_prop_kind = is_test ? dnnl::prop_kind::forward_inference
+                                       : dnnl::prop_kind::forward_training;
+    this->AcquireForwardPrimitiveDescriptor(fwd_prop_kind, x->mem_desc(),
+                                            epsilon, flags);
   }
 
   std::shared_ptr<dnnl::memory> AcquireScaleShiftMemory(const Tensor* scale,