[Chore] Consolidate reasoning/tool parser attributes into unified Parser in chat serving (#45548)

Signed-off-by: sfeng33 <4florafeng@gmail.com>
This commit is contained in:
Flora Feng
2026-06-15 11:16:45 -04:00
committed by GitHub
parent 588db18362
commit 0d80979644
4 changed files with 47 additions and 45 deletions
@@ -25,7 +25,7 @@ from vllm.entrypoints.serve.utils.api_utils import get_max_tokens
from vllm.inputs import EngineInput
from vllm.logger import init_logger
from vllm.outputs import RequestOutput
from vllm.reasoning import ReasoningParser
from vllm.parser.abstract_parser import Parser
from vllm.tokenizers import TokenizerLike
from vllm.utils.async_utils import merge_async_iterators
from vllm.utils.collection_utils import as_list
@@ -119,14 +119,15 @@ class OpenAIServingChatBatch(OpenAIServingChat):
for messages in request.messages
]
reasoning_parser: ReasoningParser | None = None
if self.reasoning_parser_cls:
parser: Parser | None = None
if self.parser_cls is not None:
chat_template_kwargs = self._effective_chat_template_kwargs(
single_requests[0]
)
reasoning_parser = self.reasoning_parser_cls(
parser = self.parser_cls(
tokenizer,
chat_template_kwargs=chat_template_kwargs, # type: ignore[call-arg]
None, # tools
chat_template_kwargs=chat_template_kwargs,
)
render_result = await self.render_batch_chat_request(request)
@@ -194,7 +195,7 @@ class OpenAIServingChatBatch(OpenAIServingChat):
all_conversations,
tokenizer,
request_metadata,
reasoning_parser,
parser,
)
async def chat_completion_full_generator_batch(
@@ -206,7 +207,7 @@ class OpenAIServingChatBatch(OpenAIServingChat):
all_conversations: list[list[ConversationMessage]],
tokenizer: TokenizerLike,
request_metadata: RequestResponseMetadata,
reasoning_parser: ReasoningParser | None = None,
parser: Parser | None = None,
) -> ErrorResponse | ChatCompletionResponse:
"""Handle batched (non-streaming) chat completions.
@@ -262,12 +263,12 @@ class OpenAIServingChatBatch(OpenAIServingChat):
else:
logprobs = None
if reasoning_parser:
reasoning, content = reasoning_parser.extract_reasoning(
if parser is not None:
reasoning, content, _ = parser.parse(
output.text,
request=request, # type: ignore[arg-type]
)
if not getattr(request, "include_reasoning", True):
if not request.include_reasoning:
reasoning = None
else:
reasoning = None