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 7, 00:15 UTC
[Deprecation][1/2] Remove items deprecated in v0.18 (#36470)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Mar 9, 2026 at 18:43 UTC
f96c3ab08cc75f18d40892ef59b6f295e71ffe83
7 files changed
+6
-76
docs/design/plugin_system.md
+1
-1
@@ -155,4 +155,4 @@ The interface for the model/module may change during vLLM's development. If you
155
- `use_v1` parameter in `Platform.get_attn_backend_cls` is deprecated. It has been removed in v0.13.0.
156
- `_Backend` in `vllm.attention` is deprecated. It has been removed in v0.13.0. Please use `vllm.v1.attention.backends.registry.register_backend` to add new attention backend to `AttentionBackendEnum` instead.
157
- `seed_everything` platform interface is deprecated. It has been removed in v0.16.0. Please use `vllm.utils.torch_utils.set_random_seed` instead.
158
- - `prompt` in `Platform.validate_request` is deprecated and will be removed in v0.18.0.
158
+ - `prompt` in `Platform.validate_request` is deprecated. It has been removed in v0.18.0.
tests/transformers_utils/test_config.py
+1
-1
@@ -3,7 +3,7 @@
3
"""
4
This test file includes some cases where it is inappropriate to
5
only get the `eos_token_id` from the tokenizer as defined by
6
-`vllm.LLMEngine._get_eos_token_id`.
6
+`BaseRenderer.get_eos_token_id`.
7
"""
8
9
from vllm.tokenizers import get_tokenizer
vllm/multimodal/processing/processor.py
-8
@@ -986,14 +986,6 @@ class BaseMultiModalProcessor(ABC, Generic[_I]):
986
self.dummy_inputs = dummy_inputs
987
self.cache = cache
988
989
- # TODO: Remove in v0.18
990
- if hasattr(self, "_get_data_parser"):
991
- raise ValueError(
992
- "BaseMultiModalProcessor._get_data_parser has been "
993
- "moved to `BaseProcessingInfo.build_data_parser` in v0.16. "
994
- "You should override `BaseProcessingInfo.build_data_parser` instead."
995
- )
996
-
989
self.data_parser = self.info.get_data_parser()
990
991
def __call__(
vllm/plugins/io_processors/__init__.py
+2
-13
@@ -1,7 +1,6 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
-import inspect
4
import logging
5
6
from vllm.config import VllmConfig
@@ -67,16 +66,6 @@ def get_io_processor(
66
f"Available plugins: {list(loadable_plugins.keys())}"
67
)
68
70
- activated_plugin_cls = loadable_plugins[model_plugin]
69
+ activated_plugin_cls = resolve_obj_by_qualname(loadable_plugins[model_plugin])
70
72
- activated_plugin_typ = resolve_obj_by_qualname(activated_plugin_cls)
73
-
74
- # for backward compatibility, the plugin does not have a renderer argument
75
- if "renderer" not in inspect.signature(activated_plugin_typ.__init__).parameters:
76
- logger.warning(
77
- "The renderer argument will be required in v0.18, "
78
- "please update your IOProcessor plugin: %s",
79
- activated_plugin_cls,
80
- )
81
- return activated_plugin_typ(vllm_config)
82
- return activated_plugin_typ(vllm_config, renderer)
71
+ return activated_plugin_cls(vllm_config, renderer)
vllm/v1/engine/__init__.py
-12
@@ -9,7 +9,6 @@ from typing import Any, Literal
9
import msgspec
10
import numpy as np
11
import torch
12
-from typing_extensions import deprecated
12
13
from vllm.lora.request import LoRARequest
14
from vllm.multimodal.inputs import MultiModalFeatureSpec
@@ -110,17 +109,6 @@ class EngineCoreRequest(
109
assert self.pooling_params is not None
110
return self.pooling_params
111
113
- @property
114
- @deprecated(
115
- "EngineCoreRequest.eos_token_id will be removed in v0.18. "
116
- "Please use EngineCoreRequest.sampling_params.eos_token_id instead."
117
- )
118
- def eos_token_id(self) -> int | None:
119
- if self.sampling_params is None:
120
- return None
121
-
122
- return self.sampling_params.eos_token_id
123
-
112
113
class EngineCoreEventType(enum.IntEnum):
114
"""The type of engine core request event."""
vllm/v1/engine/input_processor.py
+2
-29
@@ -22,13 +22,13 @@ from vllm.multimodal.inputs import (
22
MultiModalFeatureSpec,
23
)
24
from vllm.multimodal.utils import argsort_mm_positions
25
+from vllm.platforms import current_platform
26
from vllm.pooling_params import PoolingParams
27
from vllm.renderers import BaseRenderer, renderer_from_config
28
from vllm.sampling_params import SamplingParams
29
from vllm.tasks import GENERATION_TASKS, POOLING_TASKS, SupportedTask
30
from vllm.tokenizers import TokenizerLike
31
from vllm.utils import length_from_prompt_token_ids_or_embeds, random_uuid
31
-from vllm.utils.func_utils import supports_kw
32
from vllm.utils.jsontree import json_iter_leaves
33
from vllm.v1.engine import EngineCoreRequest
34
@@ -73,33 +73,6 @@ class InputProcessor:
73
mm_registry=mm_registry,
74
)
75
76
- from vllm.platforms import current_platform
77
-
78
- platform_validate_request = current_platform.validate_request
79
- if supports_kw(platform_validate_request, "prompt"):
80
- logger.warning_once(
81
- "The signature of Platform.validate_request has changed from "
82
- "`(cls, prompt, params, processed_inputs) -> None` to "
83
- "`(cls, processed_inputs, params) -> None`. The old signature "
84
- "will no longer be supported starting from v0.18."
85
- )
86
-
87
- orig_validate_request = platform_validate_request
88
-
89
- def compat_validate_request(
90
- processed_inputs: ProcessorInputs,
91
- params: SamplingParams | PoolingParams,
92
- ):
93
- return orig_validate_request(
94
- processed_inputs,
95
- params,
96
- processed_inputs, # type: ignore
97
- ) # type: ignore
98
-
99
- platform_validate_request = compat_validate_request
100
-
101
- self._platform_validate_request = platform_validate_request
102
-
76
@property
77
def tokenizer(self) -> TokenizerLike | None:
78
return self.renderer.tokenizer
@@ -265,7 +238,7 @@ class InputProcessor:
238
tokenization_kwargs=tokenization_kwargs,
239
)
240
268
- self._platform_validate_request(processed_inputs, params)
241
+ current_platform.validate_request(processed_inputs, params)
242
243
encoder_inputs, decoder_inputs = split_enc_dec_inputs(processed_inputs)
244
self._validate_model_inputs(encoder_inputs, decoder_inputs)
vllm/v1/request.py
-12
@@ -9,7 +9,6 @@ from dataclasses import dataclass
9
from typing import TYPE_CHECKING, Any
10
11
import torch
12
-from typing_extensions import deprecated
12
13
from vllm.multimodal.inputs import MultiModalFeatureSpec
14
from vllm.pooling_params import PoolingParams
@@ -177,17 +176,6 @@ class Request:
176
# None entry in the queue means finished.
177
self.streaming_queue: deque[StreamingUpdate | None] | None = None
178
180
- @property
181
- @deprecated(
182
- "Request.eos_token_id will be removed in v0.18. "
183
- "Please use Request.sampling_params.eos_token_id instead."
184
- )
185
- def eos_token_id(self) -> int | None:
186
- if self.sampling_params is None:
187
- return None
188
-
189
- return self.sampling_params.eos_token_id
190
-
179
@classmethod
180
def from_engine_core_request(
181
cls,