[Frontend] Update OpenAI error response to upstream format (#22099)

Signed-off-by: Moritz Sanft <58110325+msanft@users.noreply.github.com>
This commit is contained in:
Moritz Sanft
2025-08-06 23:06:00 -07:00
committed by GitHub
parent cbc8457b26
commit 370661856b
10 changed files with 73 additions and 67 deletions
+26 -24
View File
@@ -62,7 +62,8 @@ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
DetokenizeRequest,
DetokenizeResponse,
EmbeddingRequest,
EmbeddingResponse, ErrorResponse,
EmbeddingResponse, ErrorInfo,
ErrorResponse,
LoadLoRAAdapterRequest,
PoolingRequest, PoolingResponse,
RerankRequest, RerankResponse,
@@ -506,7 +507,7 @@ async def tokenize(request: TokenizeRequest, raw_request: Request):
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, TokenizeResponse):
return JSONResponse(content=generator.model_dump())
@@ -540,7 +541,7 @@ async def detokenize(request: DetokenizeRequest, raw_request: Request):
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, DetokenizeResponse):
return JSONResponse(content=generator.model_dump())
@@ -556,7 +557,7 @@ def maybe_register_tokenizer_info_endpoint(args):
"""Get comprehensive tokenizer information."""
result = await tokenization(raw_request).get_tokenizer_info()
return JSONResponse(content=result.model_dump(),
status_code=result.code if isinstance(
status_code=result.error.code if isinstance(
result, ErrorResponse) else 200)
@@ -603,7 +604,7 @@ async def create_responses(request: ResponsesRequest, raw_request: Request):
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, ResponsesResponse):
return JSONResponse(content=generator.model_dump())
return StreamingResponse(content=generator, media_type="text/event-stream")
@@ -620,7 +621,7 @@ async def retrieve_responses(response_id: str, raw_request: Request):
if isinstance(response, ErrorResponse):
return JSONResponse(content=response.model_dump(),
status_code=response.code)
status_code=response.error.code)
return JSONResponse(content=response.model_dump())
@@ -635,7 +636,7 @@ async def cancel_responses(response_id: str, raw_request: Request):
if isinstance(response, ErrorResponse):
return JSONResponse(content=response.model_dump(),
status_code=response.code)
status_code=response.error.code)
return JSONResponse(content=response.model_dump())
@@ -670,7 +671,7 @@ async def create_chat_completion(request: ChatCompletionRequest,
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, ChatCompletionResponse):
return JSONResponse(content=generator.model_dump())
@@ -715,7 +716,7 @@ async def create_completion(request: CompletionRequest, raw_request: Request):
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, CompletionResponse):
return JSONResponse(content=generator.model_dump())
@@ -744,7 +745,7 @@ async def create_embedding(request: EmbeddingRequest, raw_request: Request):
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, EmbeddingResponse):
return JSONResponse(content=generator.model_dump())
@@ -772,7 +773,7 @@ async def create_pooling(request: PoolingRequest, raw_request: Request):
generator = await handler.create_pooling(request, raw_request)
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, PoolingResponse):
return JSONResponse(content=generator.model_dump())
@@ -792,7 +793,7 @@ async def create_classify(request: ClassificationRequest,
generator = await handler.create_classify(request, raw_request)
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, ClassificationResponse):
return JSONResponse(content=generator.model_dump())
@@ -821,7 +822,7 @@ async def create_score(request: ScoreRequest, raw_request: Request):
generator = await handler.create_score(request, raw_request)
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, ScoreResponse):
return JSONResponse(content=generator.model_dump())
@@ -881,7 +882,7 @@ async def create_transcriptions(raw_request: Request,
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, TranscriptionResponse):
return JSONResponse(content=generator.model_dump())
@@ -922,7 +923,7 @@ async def create_translations(request: Annotated[TranslationRequest,
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, TranslationResponse):
return JSONResponse(content=generator.model_dump())
@@ -950,7 +951,7 @@ async def do_rerank(request: RerankRequest, raw_request: Request):
generator = await handler.do_rerank(request, raw_request)
if isinstance(generator, ErrorResponse):
return JSONResponse(content=generator.model_dump(),
status_code=generator.code)
status_code=generator.error.code)
elif isinstance(generator, RerankResponse):
return JSONResponse(content=generator.model_dump())
@@ -1175,7 +1176,7 @@ async def invocations(raw_request: Request):
msg = ("Cannot find suitable handler for request. "
f"Expected one of: {type_names}")
res = base(raw_request).create_error_response(message=msg)
return JSONResponse(content=res.model_dump(), status_code=res.code)
return JSONResponse(content=res.model_dump(), status_code=res.error.code)
if envs.VLLM_TORCH_PROFILER_DIR:
@@ -1211,7 +1212,7 @@ if envs.VLLM_ALLOW_RUNTIME_LORA_UPDATING:
response = await handler.load_lora_adapter(request)
if isinstance(response, ErrorResponse):
return JSONResponse(content=response.model_dump(),
status_code=response.code)
status_code=response.error.code)
return Response(status_code=200, content=response)
@@ -1223,7 +1224,7 @@ if envs.VLLM_ALLOW_RUNTIME_LORA_UPDATING:
response = await handler.unload_lora_adapter(request)
if isinstance(response, ErrorResponse):
return JSONResponse(content=response.model_dump(),
status_code=response.code)
status_code=response.error.code)
return Response(status_code=200, content=response)
@@ -1502,9 +1503,10 @@ def build_app(args: Namespace) -> FastAPI:
@app.exception_handler(HTTPException)
async def http_exception_handler(_: Request, exc: HTTPException):
err = ErrorResponse(message=exc.detail,
err = ErrorResponse(
error=ErrorInfo(message=exc.detail,
type=HTTPStatus(exc.status_code).phrase,
code=exc.status_code)
code=exc.status_code))
return JSONResponse(err.model_dump(), status_code=exc.status_code)
@app.exception_handler(RequestValidationError)
@@ -1518,9 +1520,9 @@ def build_app(args: Namespace) -> FastAPI:
else:
message = exc_str
err = ErrorResponse(message=message,
type=HTTPStatus.BAD_REQUEST.phrase,
code=HTTPStatus.BAD_REQUEST)
err = ErrorResponse(error=ErrorInfo(message=message,
type=HTTPStatus.BAD_REQUEST.phrase,
code=HTTPStatus.BAD_REQUEST))
return JSONResponse(err.model_dump(),
status_code=HTTPStatus.BAD_REQUEST)
+5 -2
View File
@@ -78,14 +78,17 @@ class OpenAIBaseModel(BaseModel):
return result
class ErrorResponse(OpenAIBaseModel):
object: str = "error"
class ErrorInfo(OpenAIBaseModel):
message: str
type: str
param: Optional[str] = None
code: int
class ErrorResponse(OpenAIBaseModel):
error: ErrorInfo
class ModelPermission(OpenAIBaseModel):
id: str = Field(default_factory=lambda: f"modelperm-{random_uuid()}")
object: str = "model_permission"
+1 -1
View File
@@ -302,7 +302,7 @@ async def run_request(serving_engine_func: Callable,
id=f"vllm-{random_uuid()}",
custom_id=request.custom_id,
response=BatchResponseData(
status_code=response.code,
status_code=response.error.code,
request_id=f"vllm-batch-{random_uuid()}"),
error=response,
)
+9 -12
View File
@@ -47,10 +47,10 @@ from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
EmbeddingChatRequest,
EmbeddingCompletionRequest,
EmbeddingRequest,
EmbeddingResponse, ErrorResponse,
PoolingResponse, RerankRequest,
ResponsesRequest, ScoreRequest,
ScoreResponse,
EmbeddingResponse, ErrorInfo,
ErrorResponse, PoolingResponse,
RerankRequest, ResponsesRequest,
ScoreRequest, ScoreResponse,
TokenizeChatRequest,
TokenizeCompletionRequest,
TokenizeResponse,
@@ -412,21 +412,18 @@ class OpenAIServing:
message: str,
err_type: str = "BadRequestError",
status_code: HTTPStatus = HTTPStatus.BAD_REQUEST) -> ErrorResponse:
return ErrorResponse(message=message,
type=err_type,
code=status_code.value)
return ErrorResponse(error=ErrorInfo(
message=message, type=err_type, code=status_code.value))
def create_streaming_error_response(
self,
message: str,
err_type: str = "BadRequestError",
status_code: HTTPStatus = HTTPStatus.BAD_REQUEST) -> str:
json_str = json.dumps({
"error":
json_str = json.dumps(
self.create_error_response(message=message,
err_type=err_type,
status_code=status_code).model_dump()
})
status_code=status_code).model_dump())
return json_str
async def _check_model(
@@ -445,7 +442,7 @@ class OpenAIServing:
if isinstance(load_result, LoRARequest):
return None
if isinstance(load_result, ErrorResponse) and \
load_result.code == HTTPStatus.BAD_REQUEST.value:
load_result.error.code == HTTPStatus.BAD_REQUEST.value:
error_response = load_result
return error_response or self.create_error_response(
+4 -5
View File
@@ -9,7 +9,7 @@ from typing import Optional, Union
from vllm.config import ModelConfig
from vllm.engine.protocol import EngineClient
from vllm.entrypoints.openai.protocol import (ErrorResponse,
from vllm.entrypoints.openai.protocol import (ErrorInfo, ErrorResponse,
LoadLoRAAdapterRequest,
ModelCard, ModelList,
ModelPermission,
@@ -82,7 +82,7 @@ class OpenAIServingModels:
load_result = await self.load_lora_adapter(
request=load_request, base_model_name=lora.base_model_name)
if isinstance(load_result, ErrorResponse):
raise ValueError(load_result.message)
raise ValueError(load_result.error.message)
def is_base_model(self, model_name) -> bool:
return any(model.name == model_name for model in self.base_model_paths)
@@ -284,6 +284,5 @@ def create_error_response(
message: str,
err_type: str = "BadRequestError",
status_code: HTTPStatus = HTTPStatus.BAD_REQUEST) -> ErrorResponse:
return ErrorResponse(message=message,
type=err_type,
code=status_code.value)
return ErrorResponse(error=ErrorInfo(
message=message, type=err_type, code=status_code.value))