diff --git a/paddle/fluid/operators/npu_op_runner.cc b/paddle/fluid/operators/npu_op_runner.cc
index ce2359db402f5bae2fad5a7e551e2e9946427679..aa0c4d2dfd274e2d834655429a600326f3af9fe2 100644
--- a/paddle/fluid/operators/npu_op_runner.cc
+++ b/paddle/fluid/operators/npu_op_runner.cc
@@ -65,9 +65,10 @@ aclFormat ConvertToNpuFormat(DataLayout layout) {
 }
 
 aclrtStream GetCurrentNPUStream() {
-  int device_id = GetCurrentNPUDeviceId();
+  int device_id = platform::GetCurrentNPUDeviceId();
   platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
-  auto &dev_ctx = *pool.Get(platform::NPUPlace(device_id));
+  auto *dev_ctx = static_cast<platform::NPUDeviceContext *>(
+      pool.Get(platform::NPUPlace(device_id)));
   return dev_ctx->stream();
 }
 
diff --git a/paddle/fluid/operators/npu_op_runner.h b/paddle/fluid/operators/npu_op_runner.h
index 81d58269513a9f71ca67d7a2269fe3cbb133a319..e178f7fc6e96d8c7cae94ac907ae383c9582303c 100644
--- a/paddle/fluid/operators/npu_op_runner.h
+++ b/paddle/fluid/operators/npu_op_runner.h
@@ -69,7 +69,7 @@ class NpuOpRunner {
 
   std::vector<aclDataBuffer *> &GetOutputBuffers();
 
-  void Run(aclrtStream stream == nullptrr);
+  void Run(aclrtStream stream = nullptr);
 
  private:
   aclTensorDesc *CreateTensorDesc(Tensor tensor);