diff --git a/cmake/third_party.cmake b/cmake/third_party.cmake
index 28b2bae9518792574860771c6049401d4c340680..8516fe8177ca55057cd8c5f21da75e1108202c30 100755
--- a/cmake/third_party.cmake
+++ b/cmake/third_party.cmake
@@ -317,8 +317,9 @@ if(WITH_ONNXRUNTIME)
 endif()
 
 if(WITH_GPU)
-  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0 OR ${CMAKE_CUDA_COMPILER_VERSION}
-                                                 GREATER_EQUAL 11.6)
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0
+     OR (${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.6
+         AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.8))
     include(external/cub) # download cub
     list(APPEND third_party_deps extern_cub)
   endif()
diff --git a/paddle/fluid/operators/fused/CMakeLists.txt b/paddle/fluid/operators/fused/CMakeLists.txt
index 23cdc33658d1c1c0b71c34fffea5551c3c496b3d..becef73569b29cb0c1ddb6493ea7757fa1ff165c 100755
--- a/paddle/fluid/operators/fused/CMakeLists.txt
+++ b/paddle/fluid/operators/fused/CMakeLists.txt
@@ -69,7 +69,7 @@ if(WITH_GPU OR WITH_ROCM)
   op_library(skip_layernorm_op)
   op_library(yolo_box_head_op)
   op_library(yolo_box_post_op)
-  op_library(fused_embedding_eltwise_layernorm_op)
+  op_library(fused_embedding_eltwise_layernorm_op DEPS bert_encoder_functor)
   op_library(fused_gate_attention_op)
   # fusion_group
   if(NOT APPLE AND NOT WIN32)