diff --git a/tensorrt_loader.py b/tensorrt_loader.py
index 3739042..6ebe51e 100644
--- a/tensorrt_loader.py
+++ b/tensorrt_loader.py
@@ -41,7 +41,6 @@ class TrTUnet:
             self.engine = runtime.deserialize_cuda_engine(f.read())
         self.context = self.engine.create_execution_context()
         self.dtype = torch.float16
-        self.stream = torch.cuda.Stream()
 
     def set_bindings_shape(self, inputs, split_batch):
         for k in inputs:
@@ -91,12 +90,13 @@ class TrTUnet:
                           dtype=trt_datatype_to_torch(self.engine.get_tensor_dtype(output_binding_name)))
         model_inputs_converted[output_binding_name] = out
 
+        stream = torch.cuda.default_stream(x.device)
         for i in range(curr_split_batch):
             for k in model_inputs_converted:
                 x = model_inputs_converted[k]
                 self.context.set_tensor_address(k, x[(x.shape[0] // curr_split_batch) * i:].data_ptr())
-            self.context.execute_async_v3(stream_handle=self.stream.cuda_stream)
-        self.stream.synchronize()
+            self.context.execute_async_v3(stream_handle=stream.cuda_stream)
+        stream.synchronize()
         return out
 
     def load_state_dict(self, sd, strict=False):