[Bugfix][Frontend] Guard remaining before-validators against non-object JSON bodies (#52528)

Signed-off-by: Kaif <kaifkohari10@gmail.com>
This commit is contained in:
Kaif Kohari
2026-08-17 02:27:59 +00:00
committed by GitHub
parent 967e104fad
commit a6a2a93f9b
7 changed files with 97 additions and 0 deletions
@@ -0,0 +1,71 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""Non-object JSON bodies must fail validation cleanly (4xx), not AttributeError (500).
mode=before validators that call data.get(...) without an isinstance(data, dict)
guard raise AttributeError for string/list/scalar bodies and surface as HTTP 500.
This extends the chat completion coverage added in #51654 to the remaining
request models whose before-validators were missing the same guard.
"""
import pytest
from pydantic import ValidationError
from vllm.entrypoints.openai.completion.protocol import CompletionRequest
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
from vllm.entrypoints.pooling.classify.protocol import ClassificationChatRequest
from vllm.entrypoints.pooling.embed.protocol import EmbeddingChatRequest
from vllm.entrypoints.pooling.pooling.protocol import PoolingChatRequest
from vllm.entrypoints.serve.tokenize.protocol import TokenizeChatRequest
from vllm.entrypoints.speech_to_text.transcription.protocol import TranscriptionRequest
from vllm.entrypoints.speech_to_text.translation.protocol import TranslationRequest
from vllm.exceptions import VLLMValidationError
pytestmark = pytest.mark.skip_global_cleanup
REQUEST_MODELS = [
CompletionRequest,
ResponsesRequest,
EmbeddingChatRequest,
ClassificationChatRequest,
PoolingChatRequest,
TokenizeChatRequest,
TranscriptionRequest,
TranslationRequest,
]
@pytest.mark.parametrize("request_model", REQUEST_MODELS, ids=lambda m: m.__name__)
@pytest.mark.parametrize(
"payload",
[
"this is not valid json{{{",
["not", "an", "object"],
42,
None,
True,
],
)
def test_request_models_reject_non_object_body(request_model, payload):
with pytest.raises(ValidationError):
request_model.model_validate(payload)
def test_completion_request_still_validates_dict_bodies():
"""The guard must not swallow real field-level errors on object bodies."""
with pytest.raises(VLLMValidationError, match="prompt"):
CompletionRequest.model_validate({"model": "qwen", "prompt": ""})
def test_tokenize_chat_request_still_validates_dict_bodies():
with pytest.raises(VLLMValidationError, match="add_generation_prompt"):
TokenizeChatRequest.model_validate(
{
"model": "qwen",
"messages": [{"role": "user", "content": "hello"}],
"continue_final_message": True,
"add_generation_prompt": True,
}
)
@@ -413,6 +413,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_response_format(cls, data): def validate_response_format(cls, data):
if not isinstance(data, dict):
return data
response_format = data.get("response_format") response_format = data.get("response_format")
if response_format is None: if response_format is None:
return data return data
@@ -444,6 +446,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def check_structured_outputs_count(cls, data): def check_structured_outputs_count(cls, data):
if not isinstance(data, dict):
return data
if data.get("structured_outputs", None) is None: if data.get("structured_outputs", None) is None:
return data return data
@@ -472,6 +476,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def check_logprobs(cls, data): def check_logprobs(cls, data):
if not isinstance(data, dict):
return data
if data.get("logprob_token_ids") and data.get("use_beam_search"): if data.get("logprob_token_ids") and data.get("use_beam_search"):
raise VLLMValidationError( raise VLLMValidationError(
"`logprob_token_ids` is not supported with beam search.", "`logprob_token_ids` is not supported with beam search.",
@@ -532,6 +538,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_stream_options(cls, data): def validate_stream_options(cls, data):
if not isinstance(data, dict):
return data
if data.get("stream_options") and not data.get("stream"): if data.get("stream_options") and not data.get("stream"):
raise VLLMValidationError( raise VLLMValidationError(
"Stream options can only be defined when `stream=True`.", "Stream options can only be defined when `stream=True`.",
@@ -543,6 +551,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_prompt_and_prompt_embeds(cls, data): def validate_prompt_and_prompt_embeds(cls, data):
if not isinstance(data, dict):
return data
prompt = data.get("prompt") prompt = data.get("prompt")
prompt_embeds = data.get("prompt_embeds") prompt_embeds = data.get("prompt_embeds")
@@ -562,6 +572,8 @@ class CompletionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_prompt_list_length(cls, data): def validate_prompt_list_length(cls, data):
if not isinstance(data, dict):
return data
max_prompts = envs.VLLM_MAX_COMPLETION_PROMPTS max_prompts = envs.VLLM_MAX_COMPLETION_PROMPTS
prompt = data.get("prompt") prompt = data.get("prompt")
@@ -467,6 +467,8 @@ class ResponsesRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_background(cls, data): def validate_background(cls, data):
if not isinstance(data, dict):
return data
if not data.get("background"): if not data.get("background"):
return data return data
if not data.get("store", True): if not data.get("store", True):
@@ -479,6 +481,8 @@ class ResponsesRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_prompt(cls, data): def validate_prompt(cls, data):
if not isinstance(data, dict):
return data
if data.get("prompt") is not None: if data.get("prompt") is not None:
raise VLLMValidationError( raise VLLMValidationError(
"prompt template is not supported", parameter="prompt" "prompt template is not supported", parameter="prompt"
@@ -499,6 +503,8 @@ class ResponsesRequest(OpenAIBaseModel):
Invalid structures are left for Pydantic to reject. Invalid structures are left for Pydantic to reject.
""" """
if not isinstance(data, dict):
return data
input_data = data.get("input") input_data = data.get("input")
# Early return for None, strings, or bytes # Early return for None, strings, or bytes
@@ -239,6 +239,8 @@ class ChatRequestOptionsMixin(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def check_generation_prompt(cls, data): def check_generation_prompt(cls, data):
if not isinstance(data, dict):
return data
if data.get("continue_final_message") and data.get("add_generation_prompt"): if data.get("continue_final_message") and data.get("add_generation_prompt"):
raise VLLMValidationError( raise VLLMValidationError(
"Cannot set both `continue_final_message` and " "Cannot set both `continue_final_message` and "
@@ -120,6 +120,8 @@ class TokenizeChatRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def check_generation_prompt(cls, data): def check_generation_prompt(cls, data):
if not isinstance(data, dict):
return data
if data.get("continue_final_message") and data.get("add_generation_prompt"): if data.get("continue_final_message") and data.get("add_generation_prompt"):
raise VLLMValidationError( raise VLLMValidationError(
"Cannot set both `continue_final_message` and " "Cannot set both `continue_final_message` and "
@@ -285,6 +285,8 @@ class TranscriptionRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_transcription_request(cls, data): def validate_transcription_request(cls, data):
if not isinstance(data, dict):
return data
if isinstance(data.get("file"), str): if isinstance(data.get("file"), str):
raise HTTPException( raise HTTPException(
status_code=HTTPStatus.UNPROCESSABLE_ENTITY, status_code=HTTPStatus.UNPROCESSABLE_ENTITY,
@@ -271,6 +271,8 @@ class TranslationRequest(OpenAIBaseModel):
@model_validator(mode="before") @model_validator(mode="before")
@classmethod @classmethod
def validate_stream_options(cls, data): def validate_stream_options(cls, data):
if not isinstance(data, dict):
return data
stream_opts = ["stream_include_usage", "stream_continuous_usage_stats"] stream_opts = ["stream_include_usage", "stream_continuous_usage_stats"]
stream = data.get("stream", False) stream = data.get("stream", False)
if any(bool(data.get(so, False)) for so in stream_opts) and not stream: if any(bool(data.get(so, False)) for so in stream_opts) and not stream: