diff --git a/onnxruntime/contrib_ops/cuda/bert/decoder_attention.cc b/onnxruntime/contrib_ops/cuda/bert/decoder_attention.cc
index 482350dd881a0..0cd0be1714471 100644
--- a/onnxruntime/contrib_ops/cuda/bert/decoder_attention.cc
+++ b/onnxruntime/contrib_ops/cuda/bert/decoder_attention.cc
@@ -85,7 +85,7 @@ Status CheckInputs(const TensorShape& query_shape,
     return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT, "q_weights shall have shape (hidden size, hidden size)");
   }
 
-  if (kv_weights_dims[0] != hidden_size || kv_weights_dims[1] != 2 * hidden_size) {
+  if (kv_weights_dims[0] != hidden_size || kv_weights_dims[1] != static_cast<int64_t>(2 * hidden_size)) {
     return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT, "kv_weights shall have shape (hidden size, 2 * hidden size)");
   }
 
diff --git a/onnxruntime/contrib_ops/cuda/transformers/generation_device_helper.cc b/onnxruntime/contrib_ops/cuda/transformers/generation_device_helper.cc
index eb0c3b528345f..d0875dbab731d 100644
--- a/onnxruntime/contrib_ops/cuda/transformers/generation_device_helper.cc
+++ b/onnxruntime/contrib_ops/cuda/transformers/generation_device_helper.cc
@@ -510,7 +510,7 @@ Status GreedySearchProcessLogits(
   const Tensor& input = next_token_scores_value.Get<Tensor>();
 
   constexpr int axis = 1;
-  const unsigned top_k = static_cast<unsigned>(1);
+  constexpr unsigned top_k = static_cast<unsigned>(1);
   constexpr bool largest = true;
   constexpr bool sorted = false;
 
diff --git a/onnxruntime/test/contrib_ops/layer_norm_test.cc b/onnxruntime/test/contrib_ops/layer_norm_test.cc
index 9c41f218a2f2f..66a8632ff2ae4 100644
--- a/onnxruntime/test/contrib_ops/layer_norm_test.cc
+++ b/onnxruntime/test/contrib_ops/layer_norm_test.cc
@@ -92,7 +92,7 @@ TEST(CudaKernelTest, LayerNorm_SmallSizeTensor) {
 
 TEST(CudaKernelTest, LayerNorm_SmallSizeTensor_IntermediateAxis) {
   const std::vector<int64_t> X_dims{4, 20, 8, 16};
-  const int64_t axis = -2;
+  constexpr int64_t axis = -2;
   TestLayerNorm(X_dims, LAYER_NORM_OP, k_epsilon_default, axis);
 }
 
@@ -108,9 +108,9 @@ TEST(CudaKernelTest, LayerNorm_LargeSizeTensor) {
 
 TEST(CudaKernelTest, LayerNorm_MidSizeTensor_NoBias) {
   std::vector<int64_t> X_dims{8, 80, 768};
-  const int64_t axis = -1;
-  const int64_t keep_dims = 1;
-  const bool no_bias = true;
+  constexpr int64_t axis = -1;
+  constexpr int64_t keep_dims = 1;
+  constexpr bool no_bias = true;
   TestLayerNorm(X_dims, LAYER_NORM_OP, k_epsilon_default, axis, keep_dims, no_bias);
 }
 
@@ -121,7 +121,7 @@ TEST(CudaKernelTest, SimplifiedLayerNorm_SmallSizeTensor) {
 
 TEST(CudaKernelTest, SimplifiedLayerNorm_SmallSizeTensor_IntermediateAxis) {
   const std::vector<int64_t> X_dims{4, 20, 8, 16};
-  const int64_t axis = -2;
+  constexpr int64_t axis = -2;
   TestLayerNorm(X_dims, SIMPLIFIED_LAYER_NORM_OP, k_epsilon_default, axis);
 }