a47e6ffe93
Signed-off-by: Chen Zhang <zhangch99@outlook.com> Co-authored-by: LiuXiaoxuanPKU <lilyliupku@gmail.com> Co-authored-by: simon-mo <xmo@berkeley.edu> Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Co-authored-by: Hongxia Yang <62075498+hongxiayang@users.noreply.github.com> Co-authored-by: Minseok Lee <47620120+minseokl@users.noreply.github.com> Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>
65 lines
2.4 KiB
Python
65 lines
2.4 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
from collections.abc import Sequence
|
|
from typing import Optional, Union
|
|
|
|
from transformers import PreTrainedTokenizerBase
|
|
|
|
from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
|
|
DeltaMessage)
|
|
from vllm.logger import init_logger
|
|
from vllm.reasoning import ReasoningParser, ReasoningParserManager
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
|
|
@ReasoningParserManager.register_module("GptOss")
|
|
class GptOssReasoningParser(ReasoningParser):
|
|
"""
|
|
Reasoning parser for GptOss model.
|
|
|
|
The GptOss model uses harmony to extract reasoning content and this parser
|
|
is only used for detecting the end of the reasoning content.
|
|
"""
|
|
|
|
def __init__(self, tokenizer: PreTrainedTokenizerBase):
|
|
super().__init__(tokenizer)
|
|
self.reasoning_end_token_ids = self.model_tokenizer.encode(
|
|
"<|start|>assistant<|channel|>final<|message|>")
|
|
|
|
def is_reasoning_end(self, input_ids: list[int]) -> bool:
|
|
end_token_ids = self.reasoning_end_token_ids
|
|
assert len(end_token_ids) > 0, "reasoning_end_token_ids is empty"
|
|
# Check if the end sequence is present in the input_ids.
|
|
# We search from the end of input_ids to find the last match.
|
|
for i in range(len(input_ids) - len(end_token_ids), -1, -1):
|
|
if input_ids[i:i + len(end_token_ids)] == end_token_ids:
|
|
return True
|
|
return False
|
|
|
|
def extract_content_ids(self, input_ids: list[int]) -> list[int]:
|
|
raise RuntimeError(
|
|
"GptOss model uses harmony to extract reasoning content. This "
|
|
"function should not be called.")
|
|
|
|
def extract_reasoning_content_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],
|
|
) -> Union[DeltaMessage, None]:
|
|
raise RuntimeError(
|
|
"GptOss model uses harmony to extract reasoning content. This "
|
|
"function should not be called.")
|
|
|
|
def extract_reasoning_content(
|
|
self, model_output: str, request: ChatCompletionRequest
|
|
) -> tuple[Optional[str], Optional[str]]:
|
|
raise RuntimeError(
|
|
"GptOss model uses harmony to extract reasoning content. This "
|
|
"function should not be called.")
|