init

vllm/transformers_utils/tokenizer_group/base_tokenizer_group.py

@@ -0,0 +1,55 @@
+from abc import ABC, abstractmethod
+from typing import List, Optional
+
+from transformers import PreTrainedTokenizer
+
+from vllm.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