From fd19a762375843c6c63fe8a110a76989fbcd1619 Mon Sep 17 00:00:00 2001
From: BingjiaWang <wangbingjia.wbj@ant-intl.com>
Date: Fri, 10 Jul 2026 17:05:26 +0800
Subject: [PATCH] [BUG] fix strip streaming empty-string suffix from DSV4 tool
 arguments (#29883)

---
 .../srt/entrypoints/openai/serving_chat.py    |  9 ++-
 .../entrypoints/openai/test_serving_chat.py   | 75 +++++++++++++++++++
 2 files changed, 81 insertions(+), 3 deletions(-)

diff --git a/python/sglang/srt/entrypoints/openai/serving_chat.py b/python/sglang/srt/entrypoints/openai/serving_chat.py
index b72dac910..bd0607682 100644
--- a/python/sglang/srt/entrypoints/openai/serving_chat.py
+++ b/python/sglang/srt/entrypoints/openai/serving_chat.py
@@ -2019,13 +2019,16 @@ class OpenAIServingChat(OpenAIServingBase):
 
         # Get expected vs actual arguments
         expected_args = detector.prev_tool_call_arr[tool_index].get("arguments", {})
-        expected_call = json.dumps(expected_args, ensure_ascii=False)
+        if isinstance(expected_args, str):
+            expected_call = expected_args
+        else:
+            expected_call = json.dumps(expected_args, ensure_ascii=False)
         actual_call = detector.streamed_args_for_tool[tool_index]
 
         # Check if there are remaining arguments to send
         remaining_call = (
-            expected_call.replace(actual_call, "", 1)
-            if actual_call in expected_call
+            expected_call[len(actual_call) :]
+            if expected_call.startswith(actual_call)
             else ""
         )
 
diff --git a/test/registered/unit/entrypoints/openai/test_serving_chat.py b/test/registered/unit/entrypoints/openai/test_serving_chat.py
index 16b38857d..8e3ba3b66 100644
--- a/test/registered/unit/entrypoints/openai/test_serving_chat.py
+++ b/test/registered/unit/entrypoints/openai/test_serving_chat.py
@@ -816,6 +816,81 @@ class ServingChatTestCase(unittest.TestCase):
         # Should return None since no completion is needed
         self.assertIsNone(result, "Should return None when no completion is needed")
 
+    def test_unstreamed_tool_args_raw_string_no_completion_needed(self):
+        """Test raw JSON string arguments are not JSON-encoded again at finish."""
+
+        mock_parser = Mock()
+        mock_detector = Mock()
+        mock_detector.prev_tool_call_arr = [
+            {"name": "report_template_recognition_commit", "arguments": "{}"}
+        ]
+        mock_detector.streamed_args_for_tool = ["{}"]
+        mock_parser.detector = mock_detector
+
+        content = {
+            "meta_info": {
+                "id": "chatcmpl-test123",
+            }
+        }
+
+        request = ChatCompletionRequest(
+            model="test",
+            messages=[{"role": "user", "content": "commit"}],
+            tools=[
+                {
+                    "type": "function",
+                    "function": {"name": "report_template_recognition_commit"},
+                }
+            ],
+        )
+
+        result = self.chat._check_for_unstreamed_tool_args(
+            parser=mock_parser,
+            content=content,
+            request=request,
+            index=0,
+        )
+
+        self.assertIsNone(result, "Should not append encoded quotes")
+
+    def test_unstreamed_tool_args_raw_string_completion(self):
+        """Test remaining raw JSON string arguments are sent at finish."""
+
+        mock_parser = Mock()
+        mock_detector = Mock()
+        mock_detector.prev_tool_call_arr = [
+            {
+                "name": "get_weather",
+                "arguments": '{"location": "San Francisco", "unit": "celsius"}',
+            }
+        ]
+        mock_detector.streamed_args_for_tool = ['{"location": "San Francisco"']
+        mock_parser.detector = mock_detector
+
+        content = {
+            "meta_info": {
+                "id": "chatcmpl-test123",
+            }
+        }
+
+        request = ChatCompletionRequest(
+            model="test",
+            messages=[{"role": "user", "content": "What's the weather?"}],
+            tools=[{"type": "function", "function": {"name": "get_weather"}}],
+        )
+
+        result = self.chat._check_for_unstreamed_tool_args(
+            parser=mock_parser,
+            content=content,
+            request=request,
+            index=0,
+        )
+
+        self.assertIsNotNone(result, "Should return chunk with remaining arguments")
+        chunk = json.loads(result[6:])
+        tool_calls = chunk["choices"][0]["delta"]["tool_calls"]
+        self.assertEqual(tool_calls[0]["function"]["arguments"], ', "unit": "celsius"}')
+
     def test_unstreamed_tool_args_no_parser_data(self):
         """Test that no completion chunk is sent when parser has no tool call data."""