[Structured Output][Reasoning] Improves decoding throughput for models using single-token reasoning endings. (#30056)

tests/reasoning/test_base_thinking_reasoning_parser.py

@@ -132,6 +132,41 @@ class TestBaseThinkingReasoningParserMethods:
             is False
         )
 
+    def test_is_reasoning_end_streaming(self, test_tokenizer):
+        """Test the is_reasoning_end_streaming method."""
+        parser = TestThinkingReasoningParser(test_tokenizer)
+        end_token_id = parser.end_token_id
+        start_token_id = parser.start_token_id
+
+        assert (
+            parser.is_reasoning_end_streaming([1, 2, end_token_id], [end_token_id])
+            is True
+        )
+        assert parser.is_reasoning_end_streaming([1, 2, 3, 4], [4]) is False
+        assert parser.is_reasoning_end_streaming([], []) is False
+        assert (
+            parser.is_reasoning_end_streaming(
+                [1, start_token_id, 2, end_token_id], [end_token_id]
+            )
+            is True
+        )
+        assert (
+            parser.is_reasoning_end_streaming([1, start_token_id, 2, 3], [3]) is False
+        )
+        assert (
+            parser.is_reasoning_end_streaming(
+                [1, start_token_id, 2, end_token_id, 2, start_token_id, 2],
+                [2],
+            )
+            is False
+        )
+        assert (
+            parser.is_reasoning_end_streaming(
+                [1, start_token_id, 2, end_token_id, 2, 2], [2]
+            )
+            is False
+        )
+
     def test_extract_content_ids(self, test_tokenizer):
         """Test the extract_content_ids method."""
         parser = TestThinkingReasoningParser(test_tokenizer)

tests/reasoning/test_deepseekv3_reasoning_parser.py

@@ -40,6 +40,7 @@ def test_identity_reasoning_parser_basic(tokenizer):
     input_tokens = tokenizer.tokenize(input_text)
     input_ids = tokenizer.convert_tokens_to_ids(input_tokens)
     assert parser.is_reasoning_end(input_ids) is True
+    assert parser.is_reasoning_end_streaming(input_ids, input_ids) is True
 
     # Test extract_content_ids returns all input_ids
     assert parser.extract_content_ids(input_ids) == input_ids