Check only end_token_id during streaming

Signed-off-by: hdlj-h <hubert@hcompany.ai>

Author: hdlj-h

tests/reasoning/test_base_thinking_reasoning_parser.py

@@ -155,8 +155,8 @@ def test_is_reasoning_end_streaming(self, test_tokenizer):
         )
         assert (
             parser.is_reasoning_end_streaming(
-                [1, start_token_id, 2, end_token_id, 2, start_token_id],
-                [end_token_id, 2, start_token_id],
+                [1, start_token_id, 2, end_token_id, 2, start_token_id, 2],
+                [2],
             )
             is False
         )

vllm/reasoning/basic_parsers.py

@@ -77,7 +77,10 @@ def is_reasoning_end(self, input_ids: list[int]) -> bool:
     def is_reasoning_end_streaming(
         self, input_ids: list[int], delta_ids: list[int]
     ) -> bool:
-        return self.is_reasoning_end(delta_ids)
+        # We only check the end token beause interleaved reasoning and content is not
+        # compatible with the current structured output manager.
+        end_token_id = self.end_token_id
+        return end_token_id in delta_ids
 
     def extract_content_ids(self, input_ids: list[int]) -> list[int]:
         """