{Deprecation] Remove tokenizer setter (#30400)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/entrypoints/llm.py

@@ -9,7 +9,7 @@ import cloudpickle
 import torch.nn as nn
 from pydantic import ValidationError
 from tqdm.auto import tqdm
-from typing_extensions import TypeVar, deprecated
+from typing_extensions import TypeVar
 
 from vllm.beam_search import (
     BeamSearchInstance,
@@ -73,7 +73,6 @@ from vllm.pooling_params import PoolingParams
 from vllm.sampling_params import BeamSearchParams, RequestOutputKind, SamplingParams
 from vllm.tasks import PoolingTask
 from vllm.tokenizers import MistralTokenizer, TokenizerLike
-from vllm.tokenizers.hf import get_cached_tokenizer
 from vllm.usage.usage_lib import UsageContext
 from vllm.utils.collection_utils import as_iter, is_list_of
 from vllm.utils.counter import Counter
@@ -367,16 +366,6 @@ class LLM:
     def get_tokenizer(self) -> TokenizerLike:
         return self.llm_engine.get_tokenizer()
 
-    @deprecated("`set_tokenizer` is deprecated and will be removed in v0.13.")
-    def set_tokenizer(self, tokenizer: TokenizerLike) -> None:
-        # While CachedTokenizer is dynamic, have no choice but
-        # compare class name. Misjudgment will arise from
-        # user-defined tokenizer started with 'Cached'
-        if tokenizer.__class__.__name__.startswith("Cached"):
-            self.llm_engine.tokenizer = tokenizer
-        else:
-            self.llm_engine.tokenizer = get_cached_tokenizer(tokenizer)
-
     def reset_mm_cache(self) -> None:
         self.input_processor.clear_mm_cache()
         self.llm_engine.reset_mm_cache()

vllm/v1/engine/async_llm.py

@@ -701,10 +701,6 @@ class AsyncLLM(EngineClient):
     def tokenizer(self) -> TokenizerLike | None:
         return self.input_processor.tokenizer
 
-    @tokenizer.setter
-    def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
-        self.input_processor.tokenizer = tokenizer
-
     async def get_tokenizer(self) -> TokenizerLike:
         if self.tokenizer is None:
             raise ValueError(

vllm/v1/engine/input_processor.py

@@ -64,10 +64,6 @@ class InputProcessor:
     def tokenizer(self) -> TokenizerLike | None:
         return self.input_preprocessor.tokenizer
 
-    @tokenizer.setter
-    def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
-        self.input_preprocessor.tokenizer = tokenizer
-
     def _validate_logprobs(
         self,
         params: SamplingParams,

vllm/v1/engine/llm_engine.py

@@ -358,10 +358,6 @@ class LLMEngine:
     def tokenizer(self) -> TokenizerLike | None:
         return self.input_processor.tokenizer
 
-    @tokenizer.setter
-    def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
-        self.input_processor.tokenizer = tokenizer
-
     def get_tokenizer(self) -> TokenizerLike:
         if self.tokenizer is None:
             raise ValueError(