diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/reduce_kernel_register.cc b/backends/iluvatar_gpu/kernels/cuda_kernels/reduce_kernel_register.cc
index ef288438bc..564666b552 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/reduce_kernel_register.cc
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/reduce_kernel_register.cc
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/reduce_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/reduce_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(reduce,
                           iluvatar_gpu,
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
index 1fd6ae7d52..40b51a4073 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/repeat_interleave_grad_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/repeat_interleave_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(repeat_interleave_with_tensor_index_grad,
                           iluvatar_gpu,
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
index e0d2421189..519e0fdf1e 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/repeat_interleave_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/repeat_interleave_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(repeat_interleave,
                           iluvatar_gpu,
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu b/backends/iluvatar_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
index 80ecfecf74..5eb1cd0c97 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/rmsprop_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/rmsprop_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(rmsprop,
                           iluvatar_gpu,
diff --git a/backends/iluvatar_gpu/kernels/cuda_kernels/roi_align_grad_kernel_register.cc b/backends/iluvatar_gpu/kernels/cuda_kernels/roi_align_grad_kernel_register.cc
index d0ca1cc48d..f3c6218246 100644
--- a/backends/iluvatar_gpu/kernels/cuda_kernels/roi_align_grad_kernel_register.cc
+++ b/backends/iluvatar_gpu/kernels/cuda_kernels/roi_align_grad_kernel_register.cc
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/roi_align_grad_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/roi_align_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(
     roi_align_grad, iluvatar_gpu, ALL_LAYOUT, phi::RoiAlignGradKernel, float) {}
diff --git a/backends/metax_gpu/CMakeLists.txt b/backends/metax_gpu/CMakeLists.txt
index af64a663c2..058215510a 100755
--- a/backends/metax_gpu/CMakeLists.txt
+++ b/backends/metax_gpu/CMakeLists.txt
@@ -124,11 +124,10 @@ 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/rmsprop_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/repeat_interleave_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/repeat_interleave_grad_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/quantize_linear_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/reduce_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/reduce_kernel_register.cu
index bccff5948d..ddc949e7f4 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/reduce_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/reduce_kernel_register.cu
@@ -14,7 +14,6 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/reduce_kernel.cu"  // NOLINT
 #include "paddle/phi/kernels/reduce_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(reduce,
diff --git a/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
index faeff6eb5e..54d09fce62 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_grad_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/repeat_interleave_grad_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/repeat_interleave_grad_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(repeat_interleave_with_tensor_index_grad,
                           metax_gpu,
diff --git a/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
index f7b20b43f5..91c7a2e9b2 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/repeat_interleave_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/repeat_interleave_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/repeat_interleave_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(repeat_interleave,
                           metax_gpu,
diff --git a/backends/metax_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu b/backends/metax_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
index 0abc2f8874..8e91aec6d4 100644
--- a/backends/metax_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
+++ b/backends/metax_gpu/kernels/cuda_kernels/rmsprop_kernel_register.cu
@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/rmsprop_kernel.cu"  // NOLINT
+#include "paddle/phi/kernels/rmsprop_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(rmsprop,
                           metax_gpu,