mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-01-20 21:14:35 +08:00
63 lines
2.1 KiB
Python
63 lines
2.1 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from collections.abc import Sequence
|
|
|
|
from transformers import PreTrainedTokenizerBase
|
|
|
|
from vllm.entrypoints.openai.protocol import ChatCompletionRequest, DeltaMessage
|
|
from vllm.logger import init_logger
|
|
from vllm.reasoning import ReasoningParser
|
|
from vllm.reasoning.deepseek_r1_reasoning_parser import DeepSeekR1ReasoningParser
|
|
|
|
from .identity_reasoning_parser import IdentityReasoningParser
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
|
|
class DeepSeekV3ReasoningParser(ReasoningParser):
|
|
"""
|
|
V3 parser that delegates to either DeepSeekR1ReasoningParser or
|
|
IdentityReasoningParser based on `thinking` and `separate_reasoning`.
|
|
"""
|
|
|
|
def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
|
|
super().__init__(tokenizer, *args, **kwargs)
|
|
|
|
chat_kwargs = kwargs.pop("chat_template_kwargs", {}) or {}
|
|
thinking = bool(chat_kwargs.pop("thinking", False))
|
|
|
|
if thinking:
|
|
self._parser = DeepSeekR1ReasoningParser(tokenizer, *args, **kwargs)
|
|
else:
|
|
self._parser = IdentityReasoningParser(tokenizer, *args, **kwargs)
|
|
|
|
def is_reasoning_end(self, input_ids: Sequence[int]) -> bool:
|
|
return self._parser.is_reasoning_end(input_ids)
|
|
|
|
def extract_content_ids(self, input_ids: list[int]) -> list[int]:
|
|
return self._parser.extract_content_ids(input_ids)
|
|
|
|
def extract_reasoning(
|
|
self, model_output: str, request: ChatCompletionRequest
|
|
) -> tuple[str | None, str | None]:
|
|
return self._parser.extract_reasoning(model_output, request)
|
|
|
|
def extract_reasoning_streaming(
|
|
self,
|
|
previous_text: str,
|
|
current_text: str,
|
|
delta_text: str,
|
|
previous_token_ids: Sequence[int],
|
|
current_token_ids: Sequence[int],
|
|
delta_token_ids: Sequence[int],
|
|
) -> DeltaMessage | None:
|
|
return self._parser.extract_reasoning_streaming(
|
|
previous_text,
|
|
current_text,
|
|
delta_text,
|
|
previous_token_ids,
|
|
current_token_ids,
|
|
delta_token_ids,
|
|
)
|