fix: LLM - Fixed the async streaming

Fixes #2853

PiperOrigin-RevId: 577345792

Author: Ark-kun

google/cloud/aiplatform/_streaming_prediction.py

@@ -130,7 +130,7 @@ async def predict_stream_of_tensor_lists_from_single_tensor_list_async(
         inputs=tensor_list,
         parameters=parameters_tensor,
     )
-    async for response in prediction_service_async_client.server_streaming_predict(
+    async for response in await prediction_service_async_client.server_streaming_predict(
         request=request
     ):
         yield response.outputs

tests/unit/aiplatform/test_language_models.py

@@ -1484,12 +1484,15 @@ async def test_text_generation_model_predict_streaming_async(self):
                 "text-bison@001"
             )
 
-        async def mock_server_streaming_predict_async(*args, **kwargs):
+        async def mock_server_streaming_predict_async_iter(*args, **kwargs):
             for response_dict in _TEST_TEXT_GENERATION_PREDICTION_STREAMING:
                 yield gca_prediction_service.StreamingPredictResponse(
                     outputs=[_streaming_prediction.value_to_tensor(response_dict)]
                 )
 
+        async def mock_server_streaming_predict_async(*args, **kwargs):
+            return mock_server_streaming_predict_async_iter(*args, **kwargs)
+
         with mock.patch.object(
             target=prediction_service_async_client.PredictionServiceAsyncClient,
             attribute="server_streaming_predict",