aphrodite-engine/aphrodite/transformers_utils/tokenizer_group/base_tokenizer_group.py

from abc import ABC, abstractmethod
from typing import List, Optional

from transformers import PreTrainedTokenizer

from aphrodite.lora.request import LoRARequest


class BaseTokenizerGroup(ABC):
    """A group of tokenizers that can be used for LoRA adapters."""

    @abstractmethod
    def ping(self) -> bool:
        """Check if the tokenizer group is alive."""
        pass

    @abstractmethod
    def get_max_input_len(self,
                          lora_request: Optional[LoRARequest] = None
                          ) -> Optional[int]:
        """Get the maximum input length for the LoRA request."""
        pass

    @abstractmethod
    def encode(self,
               prompt: str,
               request_id: Optional[str] = None,
               lora_request: Optional[LoRARequest] = None) -> List[int]:
        """Encode a prompt using the tokenizer group."""
        pass

    @abstractmethod
    async def encode_async(
            self,
            prompt: str,
            request_id: Optional[str] = None,
            lora_request: Optional[LoRARequest] = None) -> List[int]:
        """Encode a prompt using the tokenizer group."""
        pass

    @abstractmethod
    def get_lora_tokenizer(
            self,
            lora_request: Optional[LoRARequest] = None
    ) -> "PreTrainedTokenizer":
        """Get a tokenizer for a LoRA request."""
        pass

    @abstractmethod
    async def get_lora_tokenizer_async(
            self,
            lora_request: Optional[LoRARequest] = None
    ) -> "PreTrainedTokenizer":
        """Get a tokenizer for a LoRA request."""
        pass