diff --git a/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.cu b/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.cu
index a8e9a94955f70..313d58ce97ece 100644
--- a/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.cu
+++ b/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.cu
@@ -139,15 +139,9 @@ int EmbEltwiseLayernormPluginDynamicImpl<T>::enqueue(
   auto emb_ptr_gpu_d =
       emb_ptr_tensor_.mutable_data<int64_t>(platform::CUDAPlace(device_id_));
 
-  auto new_input_ptr = reinterpret_cast<uintptr_t>(inputs[0]);
-
-  if (old_input_ptr_ != new_input_ptr) {
-    old_input_ptr_ = new_input_ptr;
-
-    cudaMemcpyAsync(in_ptr_gpu_d, reinterpret_cast<const void *>(inputs),
-                    sizeof(uintptr_t) * input_num, cudaMemcpyHostToDevice,
-                    stream);
-  }
+  cudaMemcpyAsync(in_ptr_gpu_d, reinterpret_cast<const void *>(inputs),
+                  sizeof(uintptr_t) * input_num, cudaMemcpyHostToDevice,
+                  stream);
 
   auto out_type = output_desc[0].type;
 
diff --git a/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.h b/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.h
index f44391310cc21..4c9dfeef5e0dc 100644
--- a/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.h
+++ b/paddle/fluid/inference/tensorrt/plugin/emb_eltwise_layernorm_plugin.h
@@ -89,7 +89,6 @@ class EmbEltwiseLayernormPluginDynamicImpl
 
   framework::Tensor in_ptr_tensor_, emb_ptr_tensor_;
   int device_id_{0};
-  uintptr_t old_input_ptr_{0};
   bool is_initialized_{false};
 };