diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/fused_transpose_wlch_split_quant_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/fused_transpose_wlch_split_quant_kernel_register.cu
index c8145fc0e54..32144f2327e 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/fused_transpose_wlch_split_quant_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/fused_transpose_wlch_split_quant_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/fusion/gpu/fused_transpose_wlch_split_quant_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/fusion/gpu/fused_transpose_wlch_split_quant_kernel.h"
 #include "paddle/phi/kernels/fusion/gpu/quant_utils.h"
 #include "paddle/phi/kernels/primitive/datamover_primitives.h"
 
diff --git a/backends/metax_gpu/kernels/fusion/fused_transpose_wlch_split_quant_kernel_register.cu b/backends/metax_gpu/kernels/fusion/fused_transpose_wlch_split_quant_kernel_register.cu
index 574ceac4cff..ec913104de3 100644
--- a/backends/metax_gpu/kernels/fusion/fused_transpose_wlch_split_quant_kernel_register.cu
+++ b/backends/metax_gpu/kernels/fusion/fused_transpose_wlch_split_quant_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/fusion/gpu/fused_transpose_wlch_split_quant_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/fusion/gpu/fused_transpose_wlch_split_quant_kernel.h"
 #include "paddle/phi/kernels/fusion/gpu/quant_utils.h"
 #include "paddle/phi/kernels/primitive/datamover_primitives.h"