[Frontend] Migrate input-validation errors to VLLMValidationError (harmony/mistral/chat_utils/params) (#55735)

Signed-off-by: shaolila <shaolila@buaa.edu.cn>
Co-authored-by: shaolila <shaolila@buaa.edu.cn>
This commit is contained in:
Adababy
2026-09-08 06:45:40 +00:00
committed by GitHub
co-authored by shaolila
parent 755541443f
commit 5133e1d285
7 changed files with 62 additions and 8 deletions
@@ -29,6 +29,7 @@ from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionReque
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
from vllm.entrypoints.openai.responses.serving import OpenAIServingResponses
from vllm.entrypoints.serve.engine.protocol import ErrorResponse
from vllm.exceptions import VLLMValidationError
from vllm.inputs import tokens_input
from vllm.renderers.online_renderer import OnlineRenderer
from vllm.renderers.params import ChatParams
@@ -427,6 +428,19 @@ class TestReasoningRenderParity:
},
)
async def test_reasoning_effort_none_rejected_by_harmony(self, online_renderer):
"""Harmony rejects reasoning_effort="none" as invalid user input."""
request = ChatCompletionRequest(
model=_MODEL,
messages=_USER,
reasoning_effort="none",
)
with pytest.raises(
VLLMValidationError, match="Harmony does not support"
) as exc_info:
online_renderer._make_request_with_harmony(request)
assert exc_info.value.parameter == "reasoning_effort"
async def test_explicit_enable_thinking_not_overridden(
self, online_renderer, serving_responses
):
@@ -21,6 +21,7 @@ from vllm.entrypoints.chat_utils import (
_postprocess_messages,
parse_chat_messages,
parse_chat_messages_async,
validate_chat_template,
)
from vllm.exceptions import VLLMValidationError
from vllm.inputs import MultiModalDataDict, MultiModalUUIDDict
@@ -3138,3 +3139,12 @@ def test_tool_call_arguments_multiple_independent(caplog):
assert len(caplog.records) == 1
assert "bad" in caplog.records[0].message
def test_validate_chat_template_rejects_invalid_type():
"""A non-str/Path chat_template is invalid user input."""
with pytest.raises(
VLLMValidationError, match="not a valid chat template type"
) as exc_info:
validate_chat_template(123) # type: ignore[arg-type]
assert exc_info.value.parameter == "chat_template"
+14 -1
View File
@@ -11,6 +11,7 @@ from mistral_common.guidance.grammar_factory import GrammarFactory
from mistral_common.tokens.tokenizers.base import SpecialTokenPolicy, SpecialTokens
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.exceptions import VLLMValidationError
from vllm.tokenizers.mistral import (
MistralTokenizer,
_validate_apply_chat_template_args,
@@ -2279,8 +2280,20 @@ def test_validate_request_params_rejects_unsupported_effort() -> None:
messages=[],
reasoning_effort="medium",
)
with pytest.raises(ValueError, match="reasoning_effort"):
with pytest.raises(VLLMValidationError, match="reasoning_effort") as exc_info:
validate_request_params(request)
assert exc_info.value.parameter == "reasoning_effort"
def test_validate_request_params_rejects_chat_template() -> None:
request = ChatCompletionRequest(
model="test-model",
messages=[],
chat_template="{{ messages }}",
)
with pytest.raises(VLLMValidationError, match="chat_template") as exc_info:
validate_request_params(request)
assert exc_info.value.parameter == "chat_template"
# ---------------------------------------------------------------------------
+4 -1
View File
@@ -1454,7 +1454,10 @@ def validate_chat_template(chat_template: Path | str | None):
)
else:
raise TypeError(f"{type(chat_template)} is not a valid chat template type")
raise VLLMValidationError(
f"{type(chat_template)} is not a valid chat template type",
parameter="chat_template",
)
def _load_chat_template(
+4 -1
View File
@@ -512,7 +512,10 @@ class OnlineRenderer:
assert not self.supports_browsing
assert not self.supports_code_interpreter
if (reasoning_effort := request.reasoning_effort) == "none":
raise ValueError(f"Harmony does not support {reasoning_effort=}")
raise VLLMValidationError(
f"Harmony does not support {reasoning_effort=}",
parameter="reasoning_effort",
)
tools = request.tools if should_include_tools else None
messages.extend(
build_harmony_preamble(
+8 -2
View File
@@ -445,9 +445,15 @@ class TokenizeParams:
return tokens
if tokenizer is None:
raise ValueError("Cannot pad tokens when `skip_tokenizer_init=True`")
raise VLLMValidationError(
"Cannot pad tokens when `skip_tokenizer_init=True`",
parameter="pad_prompt_tokens",
)
if not isinstance(tokens, list):
raise ValueError("Cannot pad tokens for embedding inputs")
raise VLLMValidationError(
"Cannot pad tokens for embedding inputs",
parameter="pad_prompt_tokens",
)
return tokens + [tokenizer.pad_token_id] * (pad_length - len(tokens))
+8 -3
View File
@@ -34,6 +34,7 @@ from transformers.tokenization_mistral_common import MistralCommonBackend
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.exceptions import VLLMValidationError
from vllm.logger import init_logger
from vllm.tokenizers.protocol import TokenizerLike
@@ -144,15 +145,19 @@ def _validate_apply_chat_template_args(
def validate_request_params(request: "ChatCompletionRequest"):
if request.chat_template is not None or request.chat_template_kwargs is not None:
raise ValueError("chat_template is not supported for Mistral tokenizers.")
raise VLLMValidationError(
"chat_template is not supported for Mistral tokenizers.",
parameter="chat_template",
)
if request.reasoning_effort and request.reasoning_effort not in list(
ReasoningEffort
):
raise ValueError(
raise VLLMValidationError(
f"reasoning_effort={request.reasoning_effort} is not supported by "
"Mistral models. Supported values are: "
f"{[e.value for e in ReasoningEffort]}."
f"{[e.value for e in ReasoningEffort]}.",
parameter="reasoning_effort",
)