diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
index ca963664a9a..b14fd54b644 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
@@ -15,7 +15,7 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/skip_layernorm_functor.h"
-#include "paddle/phi/kernels/fusion/gpu/skip_layernorm_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/fusion/gpu/skip_layernorm_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(skip_layernorm,
                           iluvatar_gpu,
diff --git a/backends/metax_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
index b4a5c7f7155..c70c89240f3 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/skip_layernorm_kernel_register.cu
@@ -15,7 +15,7 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/skip_layernorm_functor.h"
-#include "paddle/phi/kernels/fusion/gpu/skip_layernorm_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/fusion/gpu/skip_layernorm_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(skip_layernorm,
                           metax_gpu,