refactor: migrate incremental_to_full flag from headers to user_agent parameter

Author: kevin

dashscope/aigc/generation.py

@@ -179,11 +179,13 @@ def call(  # pylint: disable=arguments-renamed,too-many-branches,too-many-statem
             to_merge_incremental_output = True
             parameters["incremental_output"] = True
 
-        # Pass incremental_to_full flag via headers user-agent
-        if "headers" not in parameters:
-            parameters["headers"] = {}
+        # Pass incremental_to_full flag via user_agent parameter
         flag = "1" if to_merge_incremental_output else "0"
-        parameters["headers"]["user-agent"] = f"incremental_to_full/{flag}"
+        existing_ua = parameters.get("user_agent", "")
+        new_ua = f"incremental_to_full/{flag}"
+        parameters["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
+        )
 
         response = super().call(
             model=model,
@@ -434,11 +436,13 @@ async def call(  # type: ignore[override] # pylint: disable=arguments-renamed,to
             to_merge_incremental_output = True
             parameters["incremental_output"] = True
 
-        # Pass incremental_to_full flag via headers user-agent
-        if "headers" not in parameters:
-            parameters["headers"] = {}
+        # Pass incremental_to_full flag via user_agent parameter
         flag = "1" if to_merge_incremental_output else "0"
-        parameters["headers"]["user-agent"] = f"incremental_to_full/{flag}"
+        existing_ua = parameters.get("user_agent", "")
+        new_ua = f"incremental_to_full/{flag}"
+        parameters["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
+        )
 
         response = await super().call(
             model=model,

dashscope/aigc/image_generation.py

@@ -99,12 +99,13 @@ def call(  # type: ignore[override]
             to_merge_incremental_output = True
             kwargs["incremental_output"] = True
 
-        # Pass incremental_to_full flag via headers user-agent
-        if "headers" not in kwargs:
-            kwargs["headers"] = {}
-
+        # Pass incremental_to_full flag via user_agent parameter
         flag = "1" if to_merge_incremental_output else "0"
-        kwargs["headers"]["user-agent"] = f"incremental_to_full/{flag}"
+        existing_ua = kwargs.get("user_agent", "")
+        new_ua = f"incremental_to_full/{flag}"
+        kwargs["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
+        )
         if kwargs.get("is_async", False):
             kwargs["headers"]["X-DashScope-Async"] = "enable"
             task = cls.async_task
@@ -416,12 +417,13 @@ async def call(  # type: ignore[override]
             to_merge_incremental_output = True
             kwargs["incremental_output"] = True
 
-        # Pass incremental_to_full flag via headers user-agent
-        if "headers" not in kwargs:
-            kwargs["headers"] = {}
-
+        # Pass incremental_to_full flag via user_agent parameter
         flag = "1" if to_merge_incremental_output else "0"
-        kwargs["headers"]["user-agent"] = f"incremental_to_full/{flag}"
+        existing_ua = kwargs.get("user_agent", "")
+        new_ua = f"incremental_to_full/{flag}"
+        kwargs["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
+        )
         if kwargs.get("is_async", False):
             kwargs["headers"]["X-DashScope-Async"] = "enable"
             task = cls.async_task

dashscope/aigc/multimodal_conversation.py

@@ -163,11 +163,13 @@ def call(  # pylint: disable=arguments-renamed,too-many-branches,too-many-statem
             to_merge_incremental_output = True
             kwargs["incremental_output"] = True
 
-        # Pass incremental_to_full flag via headers user-agent
-        if "headers" not in kwargs:
-            kwargs["headers"] = {}
+        # Pass incremental_to_full flag via user_agent parameter
         flag = "1" if to_merge_incremental_output else "0"
-        kwargs["headers"]["user-agent"] = f"incremental_to_full/{flag}"
+        existing_ua = kwargs.get("user_agent", "")
+        new_ua = f"incremental_to_full/{flag}"
+        kwargs["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
+        )
 
         response = super().call(
             model=model,

dashscope/app/application.py

@@ -160,14 +160,13 @@ def call(  # type: ignore[override]
         if to_merge_incremental_output:
             kwargs["incremental_output"] = True
 
-        # Pass incremental_to_full flag via user-agent (append)
-        if "headers" not in kwargs:
-            kwargs["headers"] = {}
+        # Pass incremental_to_full flag via user_agent parameter to avoid
+        # overwriting the default SDK user-agent
         flag = "1" if to_merge_incremental_output else "0"
-        existing_ua = kwargs["headers"].get("user-agent", "")
+        existing_ua = kwargs.get("user_agent", "")
         new_ua = f"incremental_to_full/{flag}"
-        kwargs["headers"]["user-agent"] = (
-            f"{existing_ua} {new_ua}".strip() if existing_ua else new_ua
+        kwargs["user_agent"] = (
+            f"{existing_ua}; {new_ua}".strip() if existing_ua else new_ua
         )
 
         (
@@ -273,6 +272,11 @@ def _merge_application_response(cls, response):
 
         for rsp in response:
             parsed_response = ApplicationResponse.from_api_response(rsp)
+            if parsed_response.output and not hasattr(
+                parsed_response.output,
+                "choices",
+            ):
+                parsed_response.output.choices = None
             result = merge_single_response(
                 parsed_response,
                 accumulated_data,

dashscope/audio/qwen_omni/omni_realtime.py

@@ -371,7 +371,7 @@ def end_session(self, timeout: int = 20) -> None:
         # if the server returned an error or timed out, close the connection
         if error is not None:
             self.close()
-            return
+            raise RuntimeError(f"Session ended with error: {error}")
         if not finish_success:
             self.close()
             raise TimeoutError(

dashscope/tokenizers/qwen_tokenizer.py

@@ -138,11 +138,19 @@ def _split_text(text: str, chunk_size: int = _CHUNK_SIZE) -> List[str]:
                     parts.append(piece[j : j + chunk_size])
 
         chunks: List[str] = []
+        current_chunk: List[str] = []
+        current_len = 0
         for part in parts:
-            if chunks and len(chunks[-1]) + len(part) <= chunk_size:
-                chunks[-1] += part
+            if current_len + len(part) <= chunk_size:
+                current_chunk.append(part)
+                current_len += len(part)
             else:
-                chunks.append(part)
+                if current_chunk:
+                    chunks.append("".join(current_chunk))
+                current_chunk = [part]
+                current_len = len(part)
+        if current_chunk:
+            chunks.append("".join(current_chunk))
         return chunks
 
     def decode(

tests/unit/test_conversation.py

@@ -234,5 +234,5 @@ def test_not_qwen(self, mock_server: MockServer):
         assert response.output.finish_reason == "stop"
         req = mock_server.requests.get(block=True)
         assert req["model"] == Generation.Models.dolly_12b_v2
-        assert req["parameters"] == {}
+        assert "user_agent" in req["parameters"]
         assert req["input"] == {"prompt": prompt}