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

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

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

@@ -77,6 +77,8 @@ if TYPE_CHECKING:
 
 logger = logging.getLogger(__name__)
 
+_MEDIA_CONTENT_PART_TYPES = frozenset({"image_url", "video_url", "audio_url"})
+
 
 def normalize_tool_content(role: str, content):
     """Normalize tool message content from OpenAI array format to plain string.
@@ -605,6 +607,10 @@ class OpenAIServingChat(OpenAIServingBase):
         if not request.messages:
             return "Messages cannot be empty."
 
+        media_error = self._validate_media_content(request)
+        if media_error:
+            return media_error
+
         if (
             isinstance(request.tool_choice, str)
             and request.tool_choice.lower() == "required"
@@ -658,6 +664,28 @@ class OpenAIServingChat(OpenAIServingBase):
 
         return None
 
+    def _validate_media_content(self, request: ChatCompletionRequest) -> Optional[str]:
+        if self.tokenizer_manager.model_config.is_multimodal:
+            return None
+
+        media_type = next(
+            (
+                part.type
+                for message in request.messages
+                if isinstance(message.content, list)
+                for part in message.content
+                if part.type in _MEDIA_CONTENT_PART_TYPES
+            ),
+            None,
+        )
+        if media_type is None:
+            return None
+
+        return (
+            "Model only supports text input; "
+            f"received unsupported content type '{media_type}'."
+        )
+
     def _convert_to_internal_request(
         self,
         request: ChatCompletionRequest,

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

@@ -80,6 +80,10 @@ if TYPE_CHECKING:
 logger = logging.getLogger(__name__)
 
 
+class _MediaInputValidationError(ValueError):
+    pass
+
+
 class OpenAIServingResponses(OpenAIServingChat):
     """Handler for /v1/responses requests"""
 
@@ -239,6 +243,8 @@ class OpenAIServingResponses(OpenAIServingChat):
                     processed_messages,
                 ) = await self._make_request(request, prev_response, tokenizer)
 
+        except _MediaInputValidationError as e:
+            return self.create_error_response(str(e))
         except (ValueError, TypeError, RuntimeError, jinja2.TemplateError) as e:
             logger.exception("Error in preprocessing prompt inputs")
             return self.create_error_response(f"{e} {e.__cause__}")
@@ -481,6 +487,10 @@ class OpenAIServingResponses(OpenAIServingChat):
             reasoning_effort=(request.reasoning.effort if request.reasoning else None),
         )
 
+        media_error = self._validate_media_content(chat_request)
+        if media_error:
+            raise _MediaInputValidationError(media_error)
+
         is_multimodal = self.tokenizer_manager.model_config.is_multimodal
         processed_messages = self._process_messages(chat_request, is_multimodal)