[Bugfix] Validate custom logits processor xargs for online serving (#27560)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Isotr0py
2025-11-05 16:53:33 +00:00
committed by GitHub
parent 6cae1e5332
commit 3f5a4b6473
18 changed files with 239 additions and 56 deletions
+2 -2
View File
@@ -772,10 +772,10 @@ class ChatCompletionRequest(OpenAIBaseModel):
description="KVTransfer parameters used for disaggregated serving.",
)
vllm_xargs: dict[str, str | int | float] | None = Field(
vllm_xargs: dict[str, str | int | float | list[str | int | float]] | None = Field(
default=None,
description=(
"Additional request parameters with string or "
"Additional request parameters with (list of) string or "
"numeric values, used by custom extensions."
),
)
+8
View File
@@ -68,6 +68,7 @@ from vllm.transformers_utils.tokenizers import (
validate_request_params,
)
from vllm.utils.collection_utils import as_list
from vllm.v1.sample.logits_processor import validate_logits_processors_parameters
logger = init_logger(__name__)
@@ -107,6 +108,9 @@ class OpenAIServingChat(OpenAIServing):
self.trust_request_chat_template = trust_request_chat_template
self.enable_log_outputs = enable_log_outputs
# set up logits processors
self.logits_processors = self.model_config.logits_processors
# set up reasoning parser
self.reasoning_parser = self._get_reasoning_parser(
reasoning_parser_name=reasoning_parser
@@ -291,6 +295,10 @@ class OpenAIServingChat(OpenAIServing):
self.model_config.logits_processor_pattern,
self.default_sampling_params,
)
validate_logits_processors_parameters(
self.logits_processors,
sampling_params,
)
self._log_inputs(
request_id,
@@ -36,6 +36,7 @@ from vllm.sampling_params import BeamSearchParams, SamplingParams
from vllm.transformers_utils.tokenizer import AnyTokenizer
from vllm.utils.async_utils import merge_async_iterators
from vllm.utils.collection_utils import as_list
from vllm.v1.sample.logits_processor import validate_logits_processors_parameters
logger = init_logger(__name__)
@@ -59,6 +60,10 @@ class OpenAIServingCompletion(OpenAIServing):
return_tokens_as_token_ids=return_tokens_as_token_ids,
log_error_stack=log_error_stack,
)
# set up logits processors
self.logits_processors = self.model_config.logits_processors
self.enable_prompt_tokens_details = enable_prompt_tokens_details
self.default_sampling_params = self.model_config.get_diff_sampling_param()
self.enable_force_include_usage = enable_force_include_usage
@@ -181,6 +186,10 @@ class OpenAIServingCompletion(OpenAIServing):
self.model_config.logits_processor_pattern,
self.default_sampling_params,
)
validate_logits_processors_parameters(
self.logits_processors,
sampling_params,
)
request_id_item = f"{request_id}-{i}"