[Misc] Use VLLMValidationError in pooling input validation (#51931)
Signed-off-by: frank-suwen <suwenw2@outlook.com>
This commit is contained in:
@@ -7,6 +7,7 @@ import pytest
|
|||||||
|
|
||||||
from vllm import LLM, PoolingParams
|
from vllm import LLM, PoolingParams
|
||||||
from vllm.distributed import cleanup_dist_env_and_memory
|
from vllm.distributed import cleanup_dist_env_and_memory
|
||||||
|
from vllm.exceptions import VLLMValidationError
|
||||||
|
|
||||||
MODEL_NAME = "intfloat/multilingual-e5-small"
|
MODEL_NAME = "intfloat/multilingual-e5-small"
|
||||||
|
|
||||||
@@ -61,7 +62,7 @@ def test_multiple_pooling_params(llm: LLM):
|
|||||||
assert len(PROMPTS) == len(outputs)
|
assert len(PROMPTS) == len(outputs)
|
||||||
|
|
||||||
# Exception raised, if the size of params does not match the size of prompts
|
# Exception raised, if the size of params does not match the size of prompts
|
||||||
with pytest.raises(ValueError):
|
with pytest.raises(VLLMValidationError):
|
||||||
outputs = llm.encode(
|
outputs = llm.encode(
|
||||||
PROMPTS, pooling_params=pooling_params[:3], pooling_task="embed"
|
PROMPTS, pooling_params=pooling_params[:3], pooling_task="embed"
|
||||||
)
|
)
|
||||||
|
|||||||
@@ -0,0 +1,76 @@
|
|||||||
|
# SPDX-License-Identifier: Apache-2.0
|
||||||
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||||
|
|
||||||
|
from types import SimpleNamespace
|
||||||
|
|
||||||
|
import pytest
|
||||||
|
|
||||||
|
from vllm import PoolingParams
|
||||||
|
from vllm.entrypoints.pooling.base.io_processor import PoolingIOProcessor
|
||||||
|
from vllm.entrypoints.pooling.typing import OfflineEncodeInputsContext
|
||||||
|
from vllm.exceptions import VLLMValidationError
|
||||||
|
from vllm.renderers import TokenizeParams
|
||||||
|
|
||||||
|
|
||||||
|
@pytest.fixture
|
||||||
|
def processor() -> PoolingIOProcessor:
|
||||||
|
return object.__new__(PoolingIOProcessor)
|
||||||
|
|
||||||
|
|
||||||
|
def test_rejects_untrusted_request_chat_template(processor: PoolingIOProcessor):
|
||||||
|
with pytest.raises(VLLMValidationError) as exc_info:
|
||||||
|
processor._validate_chat_template("template", None, False)
|
||||||
|
|
||||||
|
assert str(exc_info.value) == (
|
||||||
|
"Chat template is passed with request, but "
|
||||||
|
"--trust-request-chat-template is not set. "
|
||||||
|
"Refused request with untrusted chat template."
|
||||||
|
)
|
||||||
|
assert exc_info.value.parameter is None
|
||||||
|
assert exc_info.value.value is None
|
||||||
|
|
||||||
|
|
||||||
|
def test_rejects_mismatched_pooling_params(processor: PoolingIOProcessor):
|
||||||
|
with pytest.raises(VLLMValidationError) as exc_info:
|
||||||
|
processor._params_to_seq([PoolingParams()], num_requests=2)
|
||||||
|
|
||||||
|
assert str(exc_info.value) == (
|
||||||
|
"The lengths of prompts (2) and params (1) must be the same."
|
||||||
|
)
|
||||||
|
assert exc_info.value.parameter is None
|
||||||
|
assert exc_info.value.value is None
|
||||||
|
|
||||||
|
|
||||||
|
def test_rejects_mismatched_lora_requests(processor: PoolingIOProcessor):
|
||||||
|
with pytest.raises(VLLMValidationError) as exc_info:
|
||||||
|
processor._lora_request_to_seq([None], num_requests=2)
|
||||||
|
|
||||||
|
assert str(exc_info.value) == (
|
||||||
|
"The lengths of prompts (2) and lora_request (1) must be the same."
|
||||||
|
)
|
||||||
|
assert exc_info.value.parameter is None
|
||||||
|
assert exc_info.value.value is None
|
||||||
|
|
||||||
|
|
||||||
|
def test_rejects_conflicting_pooling_task(processor: PoolingIOProcessor):
|
||||||
|
processor.model_config = SimpleNamespace(is_encoder_decoder=False)
|
||||||
|
processor.renderer = SimpleNamespace(
|
||||||
|
default_cmpl_tok_params=TokenizeParams(max_total_tokens=None)
|
||||||
|
)
|
||||||
|
ctx = OfflineEncodeInputsContext(
|
||||||
|
pooling_task="embed",
|
||||||
|
tokenization_kwargs=None,
|
||||||
|
lora_request=None,
|
||||||
|
priorities=None,
|
||||||
|
prompts=[[1]],
|
||||||
|
pooling_params=PoolingParams(task="classify"),
|
||||||
|
)
|
||||||
|
|
||||||
|
with pytest.raises(VLLMValidationError) as exc_info:
|
||||||
|
processor.get_request_factory_offline(ctx)
|
||||||
|
|
||||||
|
assert str(exc_info.value) == (
|
||||||
|
"You cannot overwrite param.task='classify' with pooling_task='embed'!"
|
||||||
|
)
|
||||||
|
assert exc_info.value.parameter is None
|
||||||
|
assert exc_info.value.value is None
|
||||||
@@ -13,6 +13,7 @@ from vllm.config import VllmConfig
|
|||||||
from vllm.entrypoints.chat_utils import (
|
from vllm.entrypoints.chat_utils import (
|
||||||
ChatTemplateConfig,
|
ChatTemplateConfig,
|
||||||
)
|
)
|
||||||
|
from vllm.exceptions import VLLMValidationError
|
||||||
from vllm.lora.request import LoRARequest
|
from vllm.lora.request import LoRARequest
|
||||||
from vllm.renderers import BaseRenderer, merge_kwargs
|
from vllm.renderers import BaseRenderer, merge_kwargs
|
||||||
from vllm.renderers.inputs.preprocess import parse_model_prompt, prompt_to_seq
|
from vllm.renderers.inputs.preprocess import parse_model_prompt, prompt_to_seq
|
||||||
@@ -217,7 +218,7 @@ class PoolingIOProcessor:
|
|||||||
pass
|
pass
|
||||||
elif param.task != pooling_task:
|
elif param.task != pooling_task:
|
||||||
msg = f"You cannot overwrite {param.task=!r} with {pooling_task=!r}!"
|
msg = f"You cannot overwrite {param.task=!r} with {pooling_task=!r}!"
|
||||||
raise ValueError(msg)
|
raise VLLMValidationError(msg)
|
||||||
|
|
||||||
seq_lora_requests = self._lora_request_to_seq(ctx.lora_request, num_requests)
|
seq_lora_requests = self._lora_request_to_seq(ctx.lora_request, num_requests)
|
||||||
seq_priority = self._priority_to_seq(ctx.priorities, num_requests)
|
seq_priority = self._priority_to_seq(ctx.priorities, num_requests)
|
||||||
@@ -293,7 +294,7 @@ class PoolingIOProcessor:
|
|||||||
and chat_template_kwargs.get("chat_template") is not None
|
and chat_template_kwargs.get("chat_template") is not None
|
||||||
)
|
)
|
||||||
):
|
):
|
||||||
raise ValueError(
|
raise VLLMValidationError(
|
||||||
"Chat template is passed with request, but "
|
"Chat template is passed with request, but "
|
||||||
"--trust-request-chat-template is not set. "
|
"--trust-request-chat-template is not set. "
|
||||||
"Refused request with untrusted chat template."
|
"Refused request with untrusted chat template."
|
||||||
@@ -307,7 +308,7 @@ class PoolingIOProcessor:
|
|||||||
) -> Sequence[PoolingParams]:
|
) -> Sequence[PoolingParams]:
|
||||||
if isinstance(params, Sequence):
|
if isinstance(params, Sequence):
|
||||||
if len(params) != num_requests:
|
if len(params) != num_requests:
|
||||||
raise ValueError(
|
raise VLLMValidationError(
|
||||||
f"The lengths of prompts ({num_requests}) "
|
f"The lengths of prompts ({num_requests}) "
|
||||||
f"and params ({len(params)}) must be the same."
|
f"and params ({len(params)}) must be the same."
|
||||||
)
|
)
|
||||||
@@ -323,7 +324,7 @@ class PoolingIOProcessor:
|
|||||||
) -> Sequence[LoRARequest | None]:
|
) -> Sequence[LoRARequest | None]:
|
||||||
if isinstance(lora_request, Sequence):
|
if isinstance(lora_request, Sequence):
|
||||||
if len(lora_request) != num_requests:
|
if len(lora_request) != num_requests:
|
||||||
raise ValueError(
|
raise VLLMValidationError(
|
||||||
f"The lengths of prompts ({num_requests}) "
|
f"The lengths of prompts ({num_requests}) "
|
||||||
f"and lora_request ({len(lora_request)}) must be the same."
|
f"and lora_request ({len(lora_request)}) must be the same."
|
||||||
)
|
)
|
||||||
|
|||||||
Reference in New Issue
Block a user