[fix]reject media input for text-only models (#32914)

Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>

test/registered/unit/entrypoints/openai/test_serving_chat.py

@@ -119,6 +119,74 @@ class ServingChatTestCase(unittest.TestCase):
         self.fastapi_request = Mock(spec=Request)
         self.fastapi_request.headers = {}
 
+    def test_text_only_model_rejects_media_before_generation(self):
+        media_parts = {
+            "image_url": {
+                "type": "image_url",
+                "image_url": {"url": "https://example.com/image.png"},
+            },
+            "video_url": {
+                "type": "video_url",
+                "video_url": {"url": "https://example.com/video.mp4"},
+            },
+            "audio_url": {
+                "type": "audio_url",
+                "audio_url": {"url": "https://example.com/audio.wav"},
+            },
+        }
+
+        for media_type, media_part in media_parts.items():
+            with self.subTest(media_type=media_type):
+                request = ChatCompletionRequest(
+                    model="x",
+                    messages=[
+                        {
+                            "role": "user",
+                            "content": [
+                                {"type": "text", "text": "describe"},
+                                media_part,
+                            ],
+                        }
+                    ],
+                )
+                response = get_or_create_event_loop().run_until_complete(
+                    self.chat.handle_request(request, self.fastapi_request)
+                )
+                error = json.loads(response.body)
+                self.assertEqual(response.status_code, HTTPStatus.BAD_REQUEST)
+                self.assertEqual(error["type"], "BadRequestError")
+                self.assertIn(media_type, error["message"])
+        self.tm.generate_request.assert_not_called()
+
+    def test_media_validation_does_not_reject_supported_content(self):
+        text_request = ChatCompletionRequest(
+            model="x",
+            messages=[
+                {
+                    "role": "user",
+                    "content": [{"type": "tool_reference", "name": "get_weather"}],
+                }
+            ],
+        )
+        self.assertIsNone(self.chat._validate_request(text_request))
+
+        self.tm.model_config.is_multimodal = True
+        multimodal_request = ChatCompletionRequest(
+            model="x",
+            messages=[
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "image_url",
+                            "image_url": {"url": "https://example.com/image.png"},
+                        }
+                    ],
+                }
+            ],
+        )
+        self.assertIsNone(self.chat._validate_request(multimodal_request))
+
     # ------------- conversion tests -------------
     def test_convert_to_internal_request_single(self):
         with (

test/registered/unit/entrypoints/openai/test_serving_responses.py

@@ -303,6 +303,33 @@ class FullResponseUsageTestCase(unittest.TestCase):
 
 
 class MultimodalRequestTestCase(unittest.TestCase):
+    def test_text_only_create_responses_rejects_media_before_generation(self):
+        serving = make_serving()
+        serving._process_messages = Mock()
+        request = ResponsesRequest(
+            model="x",
+            input=[
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "input_text", "text": "describe it"},
+                        {
+                            "type": "input_image",
+                            "image_url": "http://example.com/cat.png",
+                        },
+                    ],
+                }
+            ],
+            store=False,
+        )
+
+        response = asyncio.run(serving.create_responses(request))
+
+        self.assertEqual(response.status_code, 400)
+        self.assertIn(b"received unsupported content type 'image_url'", response.body)
+        serving._process_messages.assert_not_called()
+        serving.tokenizer_manager.generate_request.assert_not_called()
+
     def test_multimodal_create_responses_sends_text_and_media_to_engine(self):
         serving = make_serving(is_multimodal=True)
         captured = {}