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 12, 11:00 UTC
[Deprecation] Advance deprecation status (#29617)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Nov 28, 2025 at 03:04 UTC
a24ea5414bc5b623cde301c8c6e1c5082ecfe412
4 files changed
+3
-84
vllm/config/scheduler.py
+1
-14
@@ -7,7 +7,7 @@ from typing import TYPE_CHECKING, Any, ClassVar, Literal, cast
7
8
from pydantic import Field, field_validator
9
from pydantic.dataclasses import dataclass
10
-from typing_extensions import Self, deprecated
10
+from typing_extensions import Self
11
12
from vllm.config.utils import config
13
from vllm.logger import init_logger
@@ -224,19 +224,6 @@ class SchedulerConfig:
224
225
self.verify_max_model_len(max_model_len)
226
227
- @property
228
- @deprecated(
229
- "`SchedulerConfig.chunked_prefill_enabled` has been renamed to "
230
- "`SchedulerConfig.enable_chunked_prefill`. "
231
- "The old name will be removed in v0.12."
232
- )
233
- def chunked_prefill_enabled(self) -> bool:
234
- return self.enable_chunked_prefill
235
-
236
- @chunked_prefill_enabled.setter
237
- def chunked_prefill_enabled(self, value: bool):
238
- self.enable_chunked_prefill = value
239
-
227
def verify_max_model_len(self, max_model_len: int) -> Self:
228
if (
229
self.max_num_batched_tokens < max_model_len
vllm/distributed/parallel_state.py
-19
@@ -41,7 +41,6 @@ import torch.distributed
41
import torch.distributed._functional_collectives as funcol
42
import torch.distributed._symmetric_memory
43
from torch.distributed import Backend, ProcessGroup
44
-from typing_extensions import deprecated
44
45
import vllm.envs as envs
46
from vllm.distributed.device_communicators.base_device_communicator import (
@@ -1078,15 +1077,6 @@ def get_tp_group() -> GroupCoordinator:
1077
return _TP
1078
1079
1081
-@deprecated(
1082
- "`get_tensor_model_parallel_group` has been replaced with "
1083
- "`get_tp_group` and may be removed after v0.12. Please use "
1084
- "`get_tp_group` instead."
1085
-)
1086
-def get_tensor_model_parallel_group():
1087
- return get_tp_group()
1088
-
1089
-
1080
_DCP: GroupCoordinator | None = None
1081
1082
@@ -1130,15 +1120,6 @@ def get_pcp_group() -> GroupCoordinator:
1120
return _PCP
1121
1122
1133
-@deprecated(
1134
- "`get_pipeline_model_parallel_group` has been replaced with "
1135
- "`get_pp_group` and may be removed in v0.12. Please use "
1136
- "`get_pp_group` instead."
1137
-)
1138
-def get_pipeline_model_parallel_group():
1139
- return get_pp_group()
1140
-
1141
-
1123
@contextmanager
1124
def graph_capture(device: torch.device):
1125
"""
vllm/model_executor/models/utils.py
-49
@@ -10,7 +10,6 @@ import torch
10
import torch.nn as nn
11
from torch.func import functional_call
12
from transformers import PretrainedConfig
13
-from typing_extensions import deprecated
13
14
from vllm.config import VllmConfig
15
from vllm.distributed import (
@@ -481,54 +480,6 @@ def _merge_multimodal_embeddings(
480
return inputs_embeds
481
482
484
-@deprecated(
485
- "`merge_multimodal_embeddings` has been replaced with "
486
- "`SupportsMultiModal.embed_input_ids` and will be "
487
- "removed in v0.12."
488
-)
489
-def merge_multimodal_embeddings(
490
- input_ids: torch.Tensor,
491
- inputs_embeds: torch.Tensor,
492
- multimodal_embeddings: NestedTensors,
493
- placeholder_token_id: int | list[int],
494
-) -> torch.Tensor:
495
- """
496
- Merge `multimodal_embeddings` into `inputs_embeds` by overwriting the
497
- positions in `inputs_embeds` corresponding to placeholder tokens in
498
- `input_ids`.
499
-
500
- `placeholder_token_id` can be a list of token ids (e.g, token ids
501
- of img_start, img_break, and img_end tokens) when needed: This means
502
- the order of these tokens in the `input_ids` MUST MATCH the order of
503
- their embeddings in `multimodal_embeddings` since we need to
504
- slice-merge instead of individually scattering.
505
-
506
- For example, if input_ids is "TTTTTSIIIBIIIBIIIETTT", where
507
- - T is text token
508
- - S is image start token
509
- - I is image embedding token
510
- - B is image break token
511
- - E is image end token.
512
-
513
- Then the image embeddings (that correspond to I's) from vision encoder
514
- must be padded with embeddings of S, B, and E in the same order of
515
- input_ids for a correct embedding merge.
516
-
517
- Note:
518
- This updates `inputs_embeds` in place.
519
- """
520
- if isinstance(placeholder_token_id, list):
521
- is_multimodal = isin_list(input_ids, placeholder_token_id)
522
- else:
523
- is_multimodal = input_ids == placeholder_token_id
524
-
525
- return _merge_multimodal_embeddings(
526
- inputs_embeds,
527
- multimodal_embeddings=multimodal_embeddings,
528
- is_multimodal=is_multimodal,
529
- )
530
-
531
-
483
def isin_list(
484
elements: torch.Tensor,
485
test_elements_list: list[int],
vllm/v1/core/sched/output.py
+2
-2
@@ -126,12 +126,12 @@ class CachedRequestData:
126
return len(self.req_ids)
127
128
@cached_property
129
- @deprecated("use resumed_req_ids field")
129
+ @deprecated("This will be removed in v0.14, use `resumed_req_ids` instead.")
130
def resumed_from_preemption(self) -> list[bool]:
131
return [req_id in self.resumed_req_ids for req_id in self.req_ids]
132
133
@cached_property
134
- @deprecated("use all_token_ids field")
134
+ @deprecated("This will be removed in v0.14, use `all_token_ids` instead.")
135
def resumed_req_token_ids(self) -> list[list[int] | None]:
136
return [
137
self.all_token_ids[req_id] if req_id in self.resumed_req_ids else None