diff --git a/onnxruntime/contrib_ops/cuda/transformers/generation_cuda_impl.cu b/onnxruntime/contrib_ops/cuda/transformers/generation_cuda_impl.cu
index faf09a6eac1e6..ae595ea1b410e 100644
--- a/onnxruntime/contrib_ops/cuda/transformers/generation_cuda_impl.cu
+++ b/onnxruntime/contrib_ops/cuda/transformers/generation_cuda_impl.cu
@@ -1,7 +1,7 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
-#include <cub/util_type.cuh>
+#include "core/providers/cuda/cu_inc/cub.cuh"
 
 #include "core/providers/cuda/cuda_common.h"
 #include "core/providers/cuda/cu_inc/common.cuh"
diff --git a/onnxruntime/core/providers/cuda/cu_inc/cub.cuh b/onnxruntime/core/providers/cuda/cu_inc/cub.cuh
index fe36ff71da33b..e88f780fd30b9 100644
--- a/onnxruntime/core/providers/cuda/cu_inc/cub.cuh
+++ b/onnxruntime/core/providers/cuda/cu_inc/cub.cuh
@@ -11,10 +11,14 @@
 #if defined(_MSC_VER)
 #pragma push_macro("__out")
 #undef __out
+// CCCL cub/config.cuh has a #pragma warning(pop) without matching push in CUDA v13.3.
+#pragma warning(push)
+#pragma warning(disable : 4193)
 #endif
 
 #include <cub/cub.cuh>
 
 #if defined(_MSC_VER)
+#pragma warning(pop)
 #pragma pop_macro("__out")
 #endif