Correctly handle concatenated output when not streaming

Author: aron

replicate/use.py

@@ -368,12 +368,18 @@ def output(self) -> O:
                 ),
             )
 
-        # For non-iterator types, wait for completion and process output
+        # For non-streaming, wait for completion and process output
         self._prediction.wait()
 
         if self._prediction.status == "failed":
             raise ModelError(self._prediction)
 
+        # Handle concatenate iterators - return joined string
+        if _has_concatenate_iterator_output_type(self._schema):
+            if isinstance(self._prediction.output, list):
+                return "".join(str(item) for item in self._prediction.output)
+            return self._prediction.output
+
         # Process output for file downloads based on schema
         return _process_output_with_schema(self._prediction.output, self._schema)
 
@@ -529,12 +535,18 @@ async def output(self) -> O:
                 ),
             )
 
-        # For non-iterator types, wait for completion and process output
+        # For non-streaming, wait for completion and process output
         await self._prediction.async_wait()
 
         if self._prediction.status == "failed":
             raise ModelError(self._prediction)
 
+        # Handle concatenate iterators - return joined string
+        if _has_concatenate_iterator_output_type(self._schema):
+            if isinstance(self._prediction.output, list):
+                return "".join(str(item) for item in self._prediction.output)
+            return self._prediction.output
+
         # Process output for file downloads based on schema
         return _process_output_with_schema(self._prediction.output, self._schema)
 

tests/test_use.py

@@ -625,6 +625,53 @@ async def async_iterator():
     assert str(result) == "['Hello', ' ', 'World']"  # str() gives list representation
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("client_mode", [ClientMode.DEFAULT, ClientMode.ASYNC])
+@respx.mock
+async def test_use_concatenate_iterator_without_streaming_returns_string(client_mode):
+    """Test that concatenate iterator models without streaming=True return final concatenated string."""
+    mock_model_endpoints(
+        versions=[
+            create_mock_version(
+                {
+                    "openapi_schema": {
+                        "components": {
+                            "schemas": {
+                                "Output": {
+                                    "type": "array",
+                                    "items": {"type": "string"},
+                                    "x-cog-array-type": "iterator",
+                                    "x-cog-array-display": "concatenate",
+                                }
+                            }
+                        }
+                    }
+                }
+            )
+        ]
+    )
+    mock_prediction_endpoints(
+        predictions=[
+            create_mock_prediction(),
+            create_mock_prediction(
+                {"status": "succeeded", "output": ["Hello", " ", "world", "!"]}
+            ),
+        ]
+    )
+
+    # Call use with "acme/hotdog-detector" WITHOUT streaming=True (default behavior)
+    hotdog_detector = replicate.use(
+        "acme/hotdog-detector", use_async=client_mode == ClientMode.ASYNC
+    )
+
+    if client_mode == ClientMode.ASYNC:
+        output = await hotdog_detector(prompt="hello world")
+    else:
+        output = hotdog_detector(prompt="hello world")
+
+    assert output == "Hello world!"
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("client_mode", [ClientMode.DEFAULT, ClientMode.ASYNC])
 @respx.mock