A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 17:30 UTC
{Deprecation] Remove tokenizer setter (#30400)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Dec 11, 2025 at 03:10 UTC
e72d65b959f759fcf56b329ecaaee7d166c012d2
4 files changed
+1
-24
vllm/entrypoints/llm.py
+1
-12
@@ -9,7 +9,7 @@ import cloudpickle
9
import torch.nn as nn
10
from pydantic import ValidationError
11
from tqdm.auto import tqdm
12
-from typing_extensions import TypeVar, deprecated
12
+from typing_extensions import TypeVar
13
14
from vllm.beam_search import (
15
BeamSearchInstance,
@@ -73,7 +73,6 @@ from vllm.pooling_params import PoolingParams
73
from vllm.sampling_params import BeamSearchParams, RequestOutputKind, SamplingParams
74
from vllm.tasks import PoolingTask
75
from vllm.tokenizers import MistralTokenizer, TokenizerLike
76
-from vllm.tokenizers.hf import get_cached_tokenizer
76
from vllm.usage.usage_lib import UsageContext
77
from vllm.utils.collection_utils import as_iter, is_list_of
78
from vllm.utils.counter import Counter
@@ -367,16 +366,6 @@ class LLM:
366
def get_tokenizer(self) -> TokenizerLike:
367
return self.llm_engine.get_tokenizer()
368
370
- @deprecated("`set_tokenizer` is deprecated and will be removed in v0.13.")
371
- def set_tokenizer(self, tokenizer: TokenizerLike) -> None:
372
- # While CachedTokenizer is dynamic, have no choice but
373
- # compare class name. Misjudgment will arise from
374
- # user-defined tokenizer started with 'Cached'
375
- if tokenizer.__class__.__name__.startswith("Cached"):
376
- self.llm_engine.tokenizer = tokenizer
377
- else:
378
- self.llm_engine.tokenizer = get_cached_tokenizer(tokenizer)
379
-
369
def reset_mm_cache(self) -> None:
370
self.input_processor.clear_mm_cache()
371
self.llm_engine.reset_mm_cache()
vllm/v1/engine/async_llm.py
-4
@@ -701,10 +701,6 @@ class AsyncLLM(EngineClient):
701
def tokenizer(self) -> TokenizerLike | None:
702
return self.input_processor.tokenizer
703
704
- @tokenizer.setter
705
- def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
706
- self.input_processor.tokenizer = tokenizer
707
-
704
async def get_tokenizer(self) -> TokenizerLike:
705
if self.tokenizer is None:
706
raise ValueError(
vllm/v1/engine/input_processor.py
-4
@@ -64,10 +64,6 @@ class InputProcessor:
64
def tokenizer(self) -> TokenizerLike | None:
65
return self.input_preprocessor.tokenizer
66
67
- @tokenizer.setter
68
- def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
69
- self.input_preprocessor.tokenizer = tokenizer
70
-
67
def _validate_logprobs(
68
self,
69
params: SamplingParams,
vllm/v1/engine/llm_engine.py
-4
@@ -358,10 +358,6 @@ class LLMEngine:
358
def tokenizer(self) -> TokenizerLike | None:
359
return self.input_processor.tokenizer
360
361
- @tokenizer.setter
362
- def tokenizer(self, tokenizer: TokenizerLike | None) -> None:
363
- self.input_processor.tokenizer = tokenizer
364
-
361
def get_tokenizer(self) -> TokenizerLike:
362
if self.tokenizer is None:
363
raise ValueError(