Feat: Supports returning expected error responses to wrong requests

python/sglang/srt/entrypoints/http_server.py

@@ -25,8 +25,11 @@
 import threading
 import time
 from http import HTTPStatus
+from json import JSONDecodeError
 from typing import AsyncIterator, Dict, Optional
 
+from pydantic import ValidationError
+
 # Fix a bug of Python threading
 setattr(threading, "_register_atexit", lambda *args, **kwargs: None)
 
@@ -57,6 +60,7 @@
 from sglang.srt.managers.tokenizer_manager import TokenizerManager
 from sglang.srt.metrics.func_timer import enable_func_timer
 from sglang.srt.openai_api.adapter import (
+    create_error_response,
     v1_batches,
     v1_cancel_batch,
     v1_chat_completions,
@@ -94,6 +98,27 @@
 )
 
 
+@app.exception_handler(ValidationError)
+@app.exception_handler(JSONDecodeError)
+async def validation_exception_handler(
+    request: Request, exc: ValidationError
+) -> Response:
+    return create_error_response(
+        message=str(exc),
+        err_type="BadRequestError",
+        status_code=HTTPStatus.BAD_REQUEST,
+    )
+
+
+@app.exception_handler(Exception)
+async def exception_handler(request: Request, exc: Exception) -> Response:
+    return create_error_response(
+        message=str(exc),
+        err_type="InternalServerError",
+        status_code=HTTPStatus.INTERNAL_SERVER_ERROR,
+    )
+
+
 # Store global states
 @dataclasses.dataclass
 class _GlobalState:

python/sglang/srt/openai_api/protocol.py

@@ -16,7 +16,7 @@
 import time
 from typing import Dict, List, Optional, Union
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 from typing_extensions import Literal
 
 
@@ -325,6 +325,15 @@ class ChatCompletionRequest(BaseModel):
     lora_path: Optional[Union[List[Optional[str]], Optional[str]]] = None
     session_params: Optional[Dict] = None
 
+    @model_validator(mode="after")
+    def validate(self):
+        # messages
+        if not self.messages:
+            raise ValueError("messages cannot be empty")
+        if self.messages[-1].role == "assistant" and len(self.messages) == 1:
+            # for https://github.com/sgl-project/sglang/issues/3579
+            raise ValueError("cannot just include messages that assistant roles")
+
 
 class FunctionResponse(BaseModel):
     """Function response."""