diff --git a/vllm/entrypoints/llm.py b/vllm/entrypoints/llm.py
index 5d5c4a1cdb77b..3fce3338503ef 100644
--- a/vllm/entrypoints/llm.py
+++ b/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()
diff --git a/vllm/v1/engine/async_llm.py b/vllm/v1/engine/async_llm.py
index fa3fb7a18895a..8eff61563ccea 100644
--- a/vllm/v1/engine/async_llm.py
+++ b/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(
diff --git a/vllm/v1/engine/input_processor.py b/vllm/v1/engine/input_processor.py
index e6a94f4e3de5d..a3c18464d3f52 100644
--- a/vllm/v1/engine/input_processor.py
+++ b/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,
diff --git a/vllm/v1/engine/llm_engine.py b/vllm/v1/engine/llm_engine.py
index 1cb206c4e004c..4422eced82fea 100644
--- a/vllm/v1/engine/llm_engine.py
+++ b/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(