diff --git a/paddle/fluid/pybind/tensor_py.h b/paddle/fluid/pybind/tensor_py.h
index 4a9dbd324c90380e784cc9457845fabd858585be..c9cad15a74bdf1077d4103c5874a2c4556b73e65 100644
--- a/paddle/fluid/pybind/tensor_py.h
+++ b/paddle/fluid/pybind/tensor_py.h
@@ -190,6 +190,7 @@ void PyCUDATensorSetFromArray(
       static_cast<const platform::CUDADeviceContext *>(pool.Get(place));
   paddle::platform::GpuMemcpyAsync(dst, array.data(), sizeof(T) * array.size(),
                                    cudaMemcpyHostToDevice, dev_ctx->stream());
+  dev_ctx->Wait();
 }
 
 template <>