Fix gpt-oss response api streaming issue (#9368)

python/sglang/srt/entrypoints/context.py

@@ -107,6 +107,8 @@ class HarmonyContext(ConversationContext):
         return self._messages
 
     def need_builtin_tool_call(self) -> bool:
+        if not self.messages:
+            return False
         last_msg = self.messages[-1]
         recipient = last_msg.recipient
         return recipient is not None and (

python/sglang/srt/entrypoints/openai/serving_responses.py

@@ -944,7 +944,7 @@ class OpenAIServingResponses(OpenAIServingChat):
                                     type="output_text",
                                     text="",
                                     annotations=[],
-                                    logprobs=[],
+                                    logprobs=None,
                                 ),
                             )
                         )
@@ -992,7 +992,7 @@ class OpenAIServingResponses(OpenAIServingChat):
                                     type="output_text",
                                     text="",
                                     annotations=[],
-                                    logprobs=[],
+                                    logprobs=None,
                                 ),
                             )
                         )