diff --git a/onnxruntime/core/optimizer/layer_norm_fusion.cc b/onnxruntime/core/optimizer/layer_norm_fusion.cc
index bf36f11521be2..d4a7457e51cbb 100644
--- a/onnxruntime/core/optimizer/layer_norm_fusion.cc
+++ b/onnxruntime/core/optimizer/layer_norm_fusion.cc
@@ -413,22 +413,33 @@ Status LayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int graph_level,
     // because SkipLayerNorm kernel, for example, has dependency on single dim size
     NodeArg* scale = nullptr;
     NodeArg* bias = nullptr;
+    std::cout << "LNF Start Changes" << std::endl;
     for (size_t i = 0; i < mul_node.MutableInputDefs().size(); i++) {
-      if (graph_utils::NodeArgIsConstant(graph, *(mul_node.MutableInputDefs()[i])) ||
-          graph_utils::IsGraphInput(graph, mul_node.MutableInputDefs()[i])) {
-        if (mul_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
-          scale = mul_node.MutableInputDefs()[i];
-        }
+      if (mul_node.MutableInputDefs()[i]->Shape() == nullptr) {
+        std::cout << "LNF Mul node is Null" << std::endl;
+        continue;
       }
+      // if (graph_utils::NodeArgIsConstant(graph, *(mul_node.MutableInputDefs()[i])) ||
+      //     graph_utils::IsGraphInput(graph, mul_node.MutableInputDefs()[i])) {
+      if (mul_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
+        std::cout << "LNF Scale set" << std::endl;
+        scale = mul_node.MutableInputDefs()[i];
+      }
+      // }
     }
 
     for (size_t i = 0; i < last_add_node.MutableInputDefs().size(); i++) {
-      if (graph_utils::NodeArgIsConstant(graph, *(last_add_node.MutableInputDefs()[i])) ||
-          graph_utils::IsGraphInput(graph, last_add_node.MutableInputDefs()[i])) {
-        if (last_add_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
-          bias = last_add_node.MutableInputDefs()[i];
-        }
+      if (last_add_node.MutableInputDefs()[i]->Shape() == nullptr) {
+        std::cout << "LNF Last add node is Null" << std::endl;
+        continue;
+      }
+      // if (graph_utils::NodeArgIsConstant(graph, *(last_add_node.MutableInputDefs()[i])) ||
+      //     graph_utils::IsGraphInput(graph, last_add_node.MutableInputDefs()[i])) {
+      if (last_add_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
+        std::cout << "LNF Bias set" << std::endl;
+        bias = last_add_node.MutableInputDefs()[i];
       }
+      // }
     }
     if (scale == nullptr || bias == nullptr) {
       continue;
@@ -666,21 +677,28 @@ Status SimplifiedLayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int gr
     // scale and bias could be multi-dims; we only support it for training at the moment
     // because SkipLayerNorm kernel, for example, has dependency on single dim size
     NodeArg* scale = nullptr;
+    std::cout << "SLNF Start Changes" << std::endl;
     for (size_t i = 0; i < mul_node.MutableInputDefs().size(); i++) {
-      if (graph_utils::NodeArgIsConstant(graph, *(mul_node.MutableInputDefs()[i])) ||
-          graph_utils::IsGraphInput(graph, mul_node.MutableInputDefs()[i])) {
+      // if (graph_utils::NodeArgIsConstant(graph, *(mul_node.MutableInputDefs()[i])) ||
+      //     graph_utils::IsGraphInput(graph, mul_node.MutableInputDefs()[i])) {
+      if (mul_node.MutableInputDefs()[i]->Shape() == nullptr) {
+        std::cout << "SLNF Mul Node Nullptr" << std::endl;
+        continue;
+      }
 #ifdef ENABLE_TRAINING_CORE
-        if (axes_values.empty() ||
-            mul_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
-          scale = mul_node.MutableInputDefs()[i];
-        }
+      std::cout << "SLNF ENABLE_TRAINING_CORE ON" << std::endl;
+      if (axes_values.empty() ||
+          mul_node.MutableInputDefs()[i]->Shape()->dim_size() == static_cast<int>(axes_values.size())) {
+        scale = mul_node.MutableInputDefs()[i];
+      }
 #else
-        // Scale must be 1d.
-        if (mul_node.MutableInputDefs()[i]->Shape()->dim_size() == 1) {
-          scale = mul_node.MutableInputDefs()[i];
-        }
-#endif
+      std::cout << "SLNF ENABLE_TRAINING_CORE OFF" << std::endl;
+      // Scale must be 1d.
+      if (mul_node.MutableInputDefs()[i]->Shape()->dim_size() == 1) {
+        scale = mul_node.MutableInputDefs()[i];
       }
+#endif
+      // }
     }
 
     if (scale == nullptr) {