diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
index 3d185e4b17..cff03b3d28 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/row_conv_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/row_conv_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
     row_conv_grad, iluvatar_gpu, ALL_LAYOUT, phi::RowConvGradKernel, float) {}
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
index 72d62b993f..018e24ba9b 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/row_conv_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/row_conv_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
     row_conv, iluvatar_gpu, ALL_LAYOUT, phi::RowConvKernel, float) {}
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/seed_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/seed_kernel_register.cu
index ea44c3eba0..706a0ab79b 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/seed_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/seed_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/seed_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/seed_kernel.h"
 #include "paddle/phi/kernels/impl/seed_kernel_impl.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
index b1b387422b..67f7bedae5 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/kernels/gpu/sequence_expand_kernel.cu"  //NOLINT
 #include "paddle/phi/kernels/impl/sequence_expand_kernel_impl.h"
+#include "paddle/phi/kernels/sequence_expand_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(sequence_expand,
                           iluvatar_gpu,
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
index cec0b99398..8f14016225 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
@@ -13,8 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/shuffle_channel.h"
-#include "paddle/phi/kernels/gpu/shuffle_channel_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/shuffle_channel_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(shuffle_channel_grad,
                           iluvatar_gpu,
diff --git a/backends/metax_gpu/CMakeLists.txt b/backends/metax_gpu/CMakeLists.txt
index 2e2f9aa0e1..4c8366c930 100755
--- a/backends/metax_gpu/CMakeLists.txt
+++ b/backends/metax_gpu/CMakeLists.txt
@@ -124,11 +124,11 @@ file(
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/funcs/math/*.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/funcs/eigen/*.cu
   # kernels/gpu
-  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_send_kernel.cu
-  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_recv_kernel.cu
-  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_concat_grad_kernel.cu
-  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_allgather_kernel.cu
-  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/ctc_align_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/shuffle_channel_grad_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/sequence_expand_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/seed_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/row_conv_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/row_conv_grad_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/yolo_box_head_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/stft_grad_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/soft_relu_grad_kernel.cu
diff --git a/backends/metax_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
index 9a68f38fab..ed8b85dbcc 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/row_conv_grad_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/row_conv_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/row_conv_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
     row_conv_grad, metax_gpu, ALL_LAYOUT, phi::RowConvGradKernel, float) {}
diff --git a/backends/metax_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
index 7be42e8b69..5832a62e8d 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/row_conv_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/row_conv_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/row_conv_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
     row_conv, metax_gpu, ALL_LAYOUT, phi::RowConvKernel, float) {}
diff --git a/backends/metax_gpu/kernels/cuda_kernels/seed_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/seed_kernel_register.cu
index 6a201e6418..c1c9c6cf85 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/seed_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/seed_kernel_register.cu
@@ -14,7 +14,7 @@
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
-#include "paddle/phi/kernels/gpu/seed_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/seed_kernel.h"
 #include "paddle/phi/kernels/impl/seed_kernel_impl.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
diff --git a/backends/metax_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
index 01397ec2f0..1b99e8c0d0 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/sequence_expand_kernel_register.cu
@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/kernels/gpu/sequence_expand_kernel.cu"  //NOLINT
 #include "paddle/phi/kernels/impl/sequence_expand_kernel_impl.h"
+#include "paddle/phi/kernels/sequence_expand_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(sequence_expand,
                           metax_gpu,
diff --git a/backends/metax_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
index 3f876b10f2..05e424f07b 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/shuffle_channel_grad_kernel_register.cu
@@ -13,8 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/shuffle_channel.h"
-#include "paddle/phi/kernels/gpu/shuffle_channel_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/shuffle_channel_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(shuffle_channel_grad,
                           metax_gpu,