Skip to content

vllm.tokenizers.registry

TokenizerRegistry

Source code in vllm/tokenizers/registry.py
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,
    ) -> "TokenizerLike":
        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)

REGISTRY class-attribute instance-attribute

REGISTRY: dict[str, tuple[str, str]] = {}

get_tokenizer staticmethod

get_tokenizer(
    tokenizer_name: str, *args, **kwargs
) -> TokenizerLike
Source code in vllm/tokenizers/registry.py
@staticmethod
def get_tokenizer(
    tokenizer_name: str,
    *args,
    **kwargs,
) -> "TokenizerLike":
    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)

register staticmethod

register(name: str, module: str, class_name: str) -> None
Source code in vllm/tokenizers/registry.py
@staticmethod
def register(name: str, module: str, class_name: str) -> None:
    TokenizerRegistry.REGISTRY[name] = (module, class_name)