[MISC] Cleanup deprecated parameters (#53559)
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai> Signed-off-by: Cyrus Leung <cyrus.tl.leung@gmail.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Jee Jee Li committed
Aug 25, 2026 at 01:19 UTC
4f686e182a3460b28df9b8e26b377a5069d519fa
12 files changed
+16
-93
tests/plugins/bge_m3_sparse_plugin/bge_m3_sparse_processor/sparse_embeddings_processor.py
+4
-6
@@ -58,13 +58,11 @@ class BgeM3SparseEmbeddingsProcessor(
58
params.dimensions = self.embed_dimensions
59
return params
60
61
- def parse_request(
62
- self, request_data: object
63
- ) -> SparseEmbeddingCompletionRequestMixin:
61
+ def parse_data(self, data: object) -> SparseEmbeddingCompletionRequestMixin:
62
# for vllm.entrypoints.llm.LLM, offline mode, calls `encode` directly.
65
- if isinstance(request_data, dict):
66
- return SparseEmbeddingCompletionRequestMixin(**request_data)
67
- raise TypeError("request_data should be a dictionary")
63
+ if isinstance(data, dict):
64
+ return SparseEmbeddingCompletionRequestMixin(**data)
65
+ raise TypeError("data should be a dictionary")
66
67
def pre_process(
68
self,
vllm/config/attention.py
+1
-1
@@ -119,7 +119,7 @@ class AttentionConfig:
119
if self.use_fp4_indexer_cache is not None:
120
logger.warning(
121
"use_fp4_indexer_cache is deprecated and will be removed in "
122
- "v0.19. Use indexer_kv_dtype instead (True -> 'mxfp4')."
122
+ "v0.29. Use indexer_kv_dtype instead (True -> 'mxfp4')."
123
)
124
if self.use_fp4_indexer_cache:
125
if self.indexer_kv_dtype not in ("auto", "mxfp4"):
vllm/entrypoints/pooling/base/io_processor.py
+1
-1
@@ -213,7 +213,7 @@ class PoolingIOProcessor:
213
if param.task is None:
214
param.task = pooling_task
215
elif pooling_task == "plugin":
216
- # `plugin` task uses io_processor.parse_request to verify inputs.
216
+ # `plugin` task uses io_processor.parse_data to verify inputs.
217
# We actually allow plugin to overwrite pooling_task.
218
pass
219
elif param.task != pooling_task:
vllm/entrypoints/pooling/offline.py
+1
-1
@@ -173,7 +173,7 @@ class PoolingOfflineMixin(OfflineInferenceMixin):
173
"Try converting the model using `--convert classify`."
174
)
175
176
- # plugin task uses io_processor.parse_request to verify inputs
176
+ # plugin task uses io_processor.parse_data to verify inputs
177
if pooling_task != "plugin" and pooling_task != self.pooling_task:
178
if pooling_task not in self.supported_tasks:
179
raise ValueError(
vllm/entrypoints/pooling/pooling/serving.py
+1
-1
@@ -74,7 +74,7 @@ class ServingPooling(PoolingBaseServing):
74
assert request.task is not None
75
pooling_task = request.task
76
77
- # plugin task uses io_processor.parse_request to verify inputs
77
+ # plugin task uses io_processor.parse_data to verify inputs
78
if pooling_task != "plugin" and pooling_task != self.pooling_task:
79
if pooling_task not in self.supported_tasks:
80
raise VLLMValidationError(
vllm/envs.py
-19
@@ -91,8 +91,6 @@ if TYPE_CHECKING:
91
VLLM_FLOAT32_MATMUL_PRECISION: Literal["highest", "high", "medium"] = "highest"
92
VLLM_BATCH_INVARIANT: bool = False
93
VLLM_TRITON_USE_TD: bool | None = None
94
- # Deprecated alias of VLLM_TRITON_USE_TD (removed in v0.25).
95
- VLLM_TRITON_ATTN_USE_TD: bool | None = None
94
VLLM_GPU_SYNC_CHECK: Literal["warn", "error"] | None = None
95
MAX_JOBS: str | None = None
96
NVCC_THREADS: str | None = None
@@ -563,19 +561,6 @@ def get_env_or_set_default(
561
logger = logging.getLogger(__name__)
562
563
566
-def _deprecated_triton_attn_use_td() -> None:
567
- """Warn that VLLM_TRITON_ATTN_USE_TD was renamed to VLLM_TRITON_USE_TD.
568
-
569
- The old name is ignored; VLLM_TRITON_USE_TD is the supported variable.
570
- """
571
- if "VLLM_TRITON_ATTN_USE_TD" in os.environ:
572
- logger.warning(
573
- "VLLM_TRITON_ATTN_USE_TD is deprecated and will be removed in "
574
- "v0.25. Use VLLM_TRITON_USE_TD instead."
575
- )
576
- return None
577
-
578
-
564
def _resolve_rust_cli_path() -> str | None:
565
"""Resolve the vllm-rs binary path.
566
@@ -651,10 +636,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
636
"VLLM_GPU_SYNC_CHECK": env_with_choices(
637
"VLLM_GPU_SYNC_CHECK", None, ["warn", "error"]
638
),
654
- # Deprecated: renamed to VLLM_TRITON_USE_TD. Kept registered so it does
655
- # not trip the unknown-env-var check; warns on use and is otherwise
656
- # ignored.
657
- "VLLM_TRITON_ATTN_USE_TD": lambda: _deprecated_triton_attn_use_td(),
639
# Maximum number of compilation jobs to run in parallel.
640
# By default this is the number of CPUs
641
"MAX_JOBS": lambda: os.getenv("MAX_JOBS", None),
vllm/model_executor/models/mistral_large_3.py
+1
-1
@@ -15,7 +15,7 @@ class MistralLarge3ForCausalLM(DeepseekV3ForCausalLM):
15
# (\A...\Z) and after substitution the resulting key always starts with
16
# "model." or "lm_head.", so no later pattern can accidentally match again.
17
hf_to_vllm_mapper = WeightsMapper(
18
- orig_to_new_regex={ # noqa: B950
18
+ orig_to_new_regex={
19
regex.compile(
20
r"\Alayers\.(\d+)\.attention_norm\.weight\Z"
21
): r"model.layers.\1.input_layernorm.weight",
vllm/multimodal/utils.py
-14
@@ -12,7 +12,6 @@ from typing import TYPE_CHECKING, Any
12
import numpy as np
13
import numpy.typing as npt
14
from PIL import Image
15
-from typing_extensions import deprecated
15
16
from vllm.inputs import MultiModalPlaceholders
17
from vllm.utils.import_utils import LazyLoader
@@ -325,19 +324,6 @@ def group_and_batch_mm_kwargs(
324
yield modality, num_items, mm_kwargs_batch
325
326
328
-@deprecated(
329
- "`group_mm_kwargs_by_modality` has been renamed to `group_and_batch_mm_kwargs`. "
330
- "The old name will be removed in v0.19."
331
-)
332
-def group_mm_kwargs_by_modality(
333
- mm_kwargs: list[tuple[str, MultiModalKwargsItem]],
334
- *,
335
- device: torch.types.Device = None,
336
- pin_memory: bool = False,
337
-) -> Generator[tuple[str, int, BatchedTensorInputs], None, None]:
338
- return group_and_batch_mm_kwargs(mm_kwargs, device=device, pin_memory=pin_memory)
339
-
340
-
327
def fetch_audio(
328
audio_url: str,
329
audio_io_kwargs: dict[str, Any] | None = None,
vllm/plugins/io_processors/interface.py
-44
@@ -1,6 +1,5 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
-import warnings
3
from abc import ABC, abstractmethod
4
from collections.abc import AsyncGenerator, Sequence
5
from typing import Generic, TypeVar
@@ -25,61 +24,18 @@ class IOProcessor(ABC, Generic[IOProcessorInput, IOProcessorOutput]):
24
self.vllm_config = vllm_config
25
26
def parse_data(self, data: object) -> IOProcessorInput:
28
- if callable(parse_request := getattr(self, "parse_request", None)):
29
- warnings.warn(
30
- "`parse_request` has been renamed to `parse_data`. "
31
- "Please update your IO Processor Plugin to use the new name. "
32
- "The old name will be removed in v0.19.",
33
- DeprecationWarning,
34
- stacklevel=2,
35
- )
36
-
37
- return parse_request(data) # type: ignore
38
-
27
raise NotImplementedError
28
29
def merge_sampling_params(
30
self,
31
params: SamplingParams | None = None,
32
) -> SamplingParams:
45
- if callable(
46
- validate_or_generate_params := getattr(
47
- self, "validate_or_generate_params", None
48
- )
49
- ):
50
- warnings.warn(
51
- "`validate_or_generate_params` has been split into "
52
- "`merge_sampling_params` and `merge_pooling_params`."
53
- "Please update your IO Processor Plugin to use the new methods. "
54
- "The old name will be removed in v0.19.",
55
- DeprecationWarning,
56
- stacklevel=2,
57
- )
58
-
59
- return validate_or_generate_params(params) # type: ignore
60
-
33
return params or SamplingParams()
34
35
def merge_pooling_params(
36
self,
37
params: PoolingParams | None = None,
38
) -> PoolingParams:
67
- if callable(
68
- validate_or_generate_params := getattr(
69
- self, "validate_or_generate_params", None
70
- )
71
- ):
72
- warnings.warn(
73
- "`validate_or_generate_params` has been split into "
74
- "`merge_sampling_params` and `merge_pooling_params`."
75
- "Please update your IO Processor Plugin to use the new methods. "
76
- "The old name will be removed in v0.19.",
77
- DeprecationWarning,
78
- stacklevel=2,
79
- )
80
-
81
- return validate_or_generate_params(params) # type: ignore
82
-
39
return params or PoolingParams(task="plugin")
40
41
@abstractmethod
vllm/pooling_params.py
+1
-1
@@ -88,7 +88,7 @@ class PoolingParams(
88
return deepcopy(self)
89
90
def verify(self, model_config: ModelConfig) -> None:
91
- # plugin task uses io_processor.parse_request to verify inputs,
91
+ # plugin task uses io_processor.parse_data to verify inputs,
92
# skipping PoolingParams verify
93
if self.task == "plugin":
94
if self.skip_reading_prefix_cache is None:
vllm/v1/engine/async_llm.py
+3
-2
@@ -339,8 +339,9 @@ class AsyncLLM(EngineClient):
339
if isinstance(prompt, EngineCoreRequest):
340
logger.warning_once(
341
"Passing EngineCoreRequest to AsyncLLM.generate() and .add_requests() "
342
- "is deprecated and will be removed in v0.18. You should instead pass "
343
- "the outputs of Renderer.render_cmpl() or Renderer.render_chat()."
342
+ "is deprecated and will be removed in the future. You should "
343
+ "instead pass the outputs of Renderer.render_cmpl() or "
344
+ "Renderer.render_chat()."
345
)
346
347
request = prompt
vllm/v1/engine/llm_engine.py
+3
-2
@@ -236,8 +236,9 @@ class LLMEngine:
236
if isinstance(prompt, EngineCoreRequest):
237
logger.warning_once(
238
"Passing EngineCoreRequest to LLMEngine.generate() and .add_requests() "
239
- "is deprecated and will be removed in v0.18. You should instead pass "
240
- "the outputs of Renderer.render_cmpl() or Renderer.render_chat()."
239
+ "is deprecated and will be removed in the future. You should "
240
+ "instead pass the outputs of Renderer.render_cmpl() or "
241
+ "Renderer.render_chat()."
242
)
243
244
request = prompt