fix(deploy): handle reasoning-only stream chunks

swift/pipelines/infer/deploy.py

@@ -150,7 +150,7 @@ def _post_process(self, request_info, response, return_cmpl_response: bool = Fal
 
         is_finished = all(response.choices[i].finish_reason for i in range(len(response.choices)))
         if 'stream' in response.__class__.__name__.lower():
-            request_info['response'] += response.choices[0].delta.content
+            request_info['response'] += response.choices[0].delta.content or ''
         else:
             request_info['response'] = response.choices[0].message.content
         if return_cmpl_response:

tests/deploy/test_deploy.py

@@ -0,0 +1,27 @@
+from types import SimpleNamespace
+
+from swift.infer_engine.protocol import ChatCompletionResponseStreamChoice, ChatCompletionStreamResponse, DeltaMessage
+from swift.pipelines.infer.deploy import SwiftDeploy
+
+
+def test_post_process_reasoning_only_stream_chunk():
+    deploy = object.__new__(SwiftDeploy)
+    deploy.args = SimpleNamespace()
+    request_info = {'response': ''}
+    response = ChatCompletionStreamResponse(
+        model='test',
+        choices=[
+            ChatCompletionResponseStreamChoice(
+                index=0,
+                delta=DeltaMessage(content=None, reasoning_content='thinking'),
+                finish_reason=None,
+            )
+        ],
+    )
+
+    processed = deploy._post_process(request_info, response)
+
+    assert processed is response
+    assert processed.choices[0].delta.content is None
+    assert processed.choices[0].delta.reasoning_content == 'thinking'
+    assert request_info['response'] == ''