diff --git a/orttraining/orttraining/test/training_ops/cuda/cross_entropy_test.cc b/orttraining/orttraining/test/training_ops/cuda/cross_entropy_test.cc
index d9800ce0e0d3e..d36f9b307ec70 100644
--- a/orttraining/orttraining/test/training_ops/cuda/cross_entropy_test.cc
+++ b/orttraining/orttraining/test/training_ops/cuda/cross_entropy_test.cc
@@ -311,11 +311,9 @@ template <typename T, typename TOut>
 static std::vector<OrtValue> RunSCELossWithEP(const char* op,
                                               int opset_version,
                                               const char* domain,
-                                              std::function<std::unique_ptr<IExecutionProvider>()>
-                                                  ep_creator,
+                                              std::function<std::unique_ptr<IExecutionProvider>()> ep_creator,
                                               const std::string& reduction,
                                               const std::int64_t ignore_index,
-                                              const double error_tolerance,
                                               const std::vector<int64_t>* X_dims,
                                               const std::vector<int64_t>* index_dims,
                                               const std::vector<int64_t>* weight_dims,
@@ -403,7 +401,7 @@ static void TestSCELoss(const char* op, int opset_version,
     cpu_fetches = RunSCELossWithEP<float, float>(
         op, opset_version, domain,
         []() -> std::unique_ptr<IExecutionProvider> { return DefaultCpuExecutionProvider(); },
-        reduction, ignore_index, error_tolerance,
+        reduction, ignore_index,
         X_dims, index_dims, weight_dims,
         Y_dims, log_prob_dims,
         X_data_temp, index_data, weight_data_temp);
@@ -411,7 +409,7 @@ static void TestSCELoss(const char* op, int opset_version,
     cpu_fetches = RunSCELossWithEP<T, float>(
         op, opset_version, domain,
         []() -> std::unique_ptr<IExecutionProvider> { return DefaultCpuExecutionProvider(); },
-        reduction, ignore_index, error_tolerance,
+        reduction, ignore_index,
         X_dims, index_dims, weight_dims,
         Y_dims, log_prob_dims,
         X_data, index_data, weight_data);
@@ -429,7 +427,7 @@ static void TestSCELoss(const char* op, int opset_version,
         return DefaultRocmExecutionProvider();
 #endif
       },
-      reduction, ignore_index, error_tolerance,
+      reduction, ignore_index,
       X_dims, index_dims, weight_dims,
       Y_dims, log_prob_dims,
       X_data, index_data, weight_data);
diff --git a/orttraining/orttraining/training_ops/cuda/nn/conv_transpose_grad.cc b/orttraining/orttraining/training_ops/cuda/nn/conv_transpose_grad.cc
index d3f5a89434a48..5d12e0ac312c0 100644
--- a/orttraining/orttraining/training_ops/cuda/nn/conv_transpose_grad.cc
+++ b/orttraining/orttraining/training_ops/cuda/nn/conv_transpose_grad.cc
@@ -53,7 +53,6 @@ Status ConvTransposeGrad<T>::ComputeInputGradient(onnxruntime::Stream* stream, c
             algo_perf.algo, workspace.get(), algo_perf.memory, &zero, args.y_tensor, args.y_data));
         return Status::OK();
       });
-  return Status::OK();
 }
 
 template <typename T>
@@ -71,7 +70,6 @@ Status ConvTransposeGrad<T>::ComputeWeightGradient(onnxruntime::Stream* stream,
             algo_perf.algo, workspace.get(), algo_perf.memory, &zero, args.w_desc, args.dw_data));
         return Status::OK();
       });
-  return Status::OK();
 }
 
 template <typename T>
diff --git a/orttraining/orttraining/training_ops/cuda/nn/layer_norm_impl.cu b/orttraining/orttraining/training_ops/cuda/nn/layer_norm_impl.cu
index 2d89ed05712e0..ad577afa06c18 100644
--- a/orttraining/orttraining/training_ops/cuda/nn/layer_norm_impl.cu
+++ b/orttraining/orttraining/training_ops/cuda/nn/layer_norm_impl.cu
@@ -30,8 +30,6 @@
 namespace onnxruntime {
 namespace cuda {
 
-using namespace onnxruntime::cuda;
-
 namespace {
   // This is the un-specialized struct.  Note that we prevent instantiation of this
   // struct by putting an undefined symbol in the function body so it won't compile.
diff --git a/orttraining/orttraining/training_ops/cuda/optimizer/lamb_impl.cu b/orttraining/orttraining/training_ops/cuda/optimizer/lamb_impl.cu
index c90809eb2fdcc..fd55f7c30ff75 100644
--- a/orttraining/orttraining/training_ops/cuda/optimizer/lamb_impl.cu
+++ b/orttraining/orttraining/training_ops/cuda/optimizer/lamb_impl.cu
@@ -619,7 +619,7 @@ CudaKernel::CudaAsyncBuffer<LambMultiTensorSyncRangeAndLock> compute_tensor_rang
 
 template <typename TIn1, typename TIn2, typename TOut1, typename TOut2, typename TBuf>
 void LambMultiTensorReductionFunctor<TIn1, TIn2, TOut1, TOut2, TBuf>::operator()(
-    cudaStream_t stream,
+    cudaStream_t /*stream*/,
     ChunkGroup<4> chunk_group,
     const CudaKernel& kernel,
     void* reduction_buffer,