mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2025-12-21 02:05:01 +08:00
156 lines
3.9 KiB
Python
156 lines
3.9 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
import importlib
|
|
from abc import ABC, abstractmethod
|
|
from typing import TYPE_CHECKING, Any
|
|
|
|
if TYPE_CHECKING:
|
|
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
|
|
|
|
|
|
class TokenizerBase(ABC):
|
|
@property
|
|
@abstractmethod
|
|
def all_special_tokens_extended(self) -> list[str]:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def all_special_tokens(self) -> list[str]:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def all_special_ids(self) -> list[int]:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def bos_token_id(self) -> int:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def eos_token_id(self) -> int:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def sep_token(self) -> str:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def pad_token(self) -> str:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def is_fast(self) -> bool:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def vocab_size(self) -> int:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def max_token_id(self) -> int:
|
|
raise NotImplementedError()
|
|
|
|
@property
|
|
@abstractmethod
|
|
def truncation_side(self) -> str:
|
|
raise NotImplementedError()
|
|
|
|
def __len__(self) -> int:
|
|
return self.vocab_size
|
|
|
|
@abstractmethod
|
|
def __call__(
|
|
self,
|
|
text: str | list[str] | list[int],
|
|
text_pair: str | None = None,
|
|
add_special_tokens: bool = False,
|
|
truncation: bool = False,
|
|
max_length: int | None = None,
|
|
):
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def get_vocab(self) -> dict[str, int]:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def get_added_vocab(self) -> dict[str, int]:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def encode_one(
|
|
self,
|
|
text: str,
|
|
truncation: bool = False,
|
|
max_length: int | None = None,
|
|
) -> list[int]:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def encode(
|
|
self,
|
|
text: str,
|
|
truncation: bool | None = None,
|
|
max_length: int | None = None,
|
|
add_special_tokens: bool | None = None,
|
|
) -> list[int]:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def apply_chat_template(
|
|
self,
|
|
messages: list["ChatCompletionMessageParam"],
|
|
tools: list[dict[str, Any]] | None = None,
|
|
**kwargs,
|
|
) -> list[int]:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def convert_tokens_to_string(self, tokens: list[str]) -> str:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def decode(self, ids: list[int] | int, skip_special_tokens: bool = True) -> str:
|
|
raise NotImplementedError()
|
|
|
|
@abstractmethod
|
|
def convert_ids_to_tokens(
|
|
self,
|
|
ids: list[int],
|
|
skip_special_tokens: bool = True,
|
|
) -> list[str]:
|
|
raise NotImplementedError()
|
|
|
|
|
|
class TokenizerRegistry:
|
|
# Tokenizer name -> (tokenizer module, tokenizer class)
|
|
REGISTRY: dict[str, tuple[str, str]] = {}
|
|
|
|
@staticmethod
|
|
def register(name: str, module: str, class_name: str) -> None:
|
|
TokenizerRegistry.REGISTRY[name] = (module, class_name)
|
|
|
|
@staticmethod
|
|
def get_tokenizer(
|
|
tokenizer_name: str,
|
|
*args,
|
|
**kwargs,
|
|
) -> TokenizerBase:
|
|
tokenizer_cls = TokenizerRegistry.REGISTRY.get(tokenizer_name)
|
|
if tokenizer_cls is None:
|
|
raise ValueError(f"Tokenizer {tokenizer_name} not found.")
|
|
|
|
tokenizer_module = importlib.import_module(tokenizer_cls[0])
|
|
class_ = getattr(tokenizer_module, tokenizer_cls[1])
|
|
return class_.from_pretrained(*args, **kwargs)
|