@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 9513290032

Fork 0

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 15, 19:45 UTC

[Misc] Upgrade to Python 3.9 typing for additional directories (#14492)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Cyrus Leung committed Mar 9, 2025 at 01:35 UTC 951329003203ba34f52b82ed12bcf91b349c2fa2

12 files changed +78 -78

pyproject.toml

-5

@@ -75,11 +75,8 @@ exclude = [

75 "vllm/distributed/**/*.py" = ["UP006", "UP035"]

76 "vllm/engine/**/*.py" = ["UP006", "UP035"]

77 "vllm/executor/**/*.py" = ["UP006", "UP035"]

78 -"vllm/inputs/**/*.py" = ["UP006", "UP035"]

79 -"vllm/logging_utils/**/*.py" = ["UP006", "UP035"]

78 "vllm/lora/**/*.py" = ["UP006", "UP035"]

79 "vllm/model_executor/**/*.py" = ["UP006", "UP035"]

82 -"vllm/multimodal/**/*.py" = ["UP006", "UP035"]

80 "vllm/platforms/**/*.py" = ["UP006", "UP035"]

81 "vllm/plugins/**/*.py" = ["UP006", "UP035"]

82 "vllm/profiler/**/*.py" = ["UP006", "UP035"]

@@ -87,9 +84,7 @@ exclude = [

84 "vllm/spec_decode/**/*.py" = ["UP006", "UP035"]

85 "vllm/transformers_utils/**/*.py" = ["UP006", "UP035"]

86 "vllm/triton_utils/**/*.py" = ["UP006", "UP035"]

90 -"vllm/usage/**/*.py" = ["UP006", "UP035"]

87 "vllm/vllm_flash_attn/**/*.py" = ["UP006", "UP035"]

92 -"vllm/assets/**/*.py" = ["UP006", "UP035"]

88 "vllm/worker/**/*.py" = ["UP006", "UP035"]

90 [tool.ruff.lint]

vllm/assets/video.py

+3 -3

@@ -2,7 +2,7 @@

3 from dataclasses import dataclass

4 from functools import lru_cache

5 -from typing import List, Literal

5 +from typing import Literal

7 import cv2

8 import numpy as np

@@ -58,7 +58,7 @@ def video_to_ndarrays(path: str, num_frames: int = -1) -> npt.NDArray:

60 def video_to_pil_images_list(path: str,

61 - num_frames: int = -1) -> List[Image.Image]:

61 + num_frames: int = -1) -> list[Image.Image]:

62 frames = video_to_ndarrays(path, num_frames)

63 return [

64 Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

@@ -72,7 +72,7 @@ class VideoAsset:

72 num_frames: int = -1

74 @property

75 - def pil_images(self) -> List[Image.Image]:

75 + def pil_images(self) -> list[Image.Image]:

76 video_path = download_video_asset(self.name)

77 ret = video_to_pil_images_list(video_path, self.num_frames)

78 return ret

vllm/inputs/data.py

+27 -27

@@ -1,9 +1,9 @@

1 # SPDX-License-Identifier: Apache-2.0

3 +from collections.abc import Iterable

4 from dataclasses import dataclass

5 from functools import cached_property

5 -from typing import (TYPE_CHECKING, Any, Dict, Generic, Iterable, List, Literal,

6 - Optional, Tuple, Union, cast)

6 +from typing import TYPE_CHECKING, Any, Generic, Literal, Optional, Union, cast

8 import torch

9 from typing_extensions import NotRequired, TypedDict, TypeVar, assert_never

@@ -26,7 +26,7 @@ class TextPrompt(TypedDict):

26 if the model supports it.

27 """

29 - mm_processor_kwargs: NotRequired[Dict[str, Any]]

29 + mm_processor_kwargs: NotRequired[dict[str, Any]]

30 """

31 Optional multi-modal processor kwargs to be forwarded to the

32 multimodal input mapper & processor. Note that if multiple modalities

@@ -38,10 +38,10 @@ class TextPrompt(TypedDict):

38 class TokensPrompt(TypedDict):

39 """Schema for a tokenized prompt."""

41 - prompt_token_ids: List[int]

41 + prompt_token_ids: list[int]

42 """A list of token IDs to pass to the model."""

44 - token_type_ids: NotRequired[List[int]]

44 + token_type_ids: NotRequired[list[int]]

45 """A list of token type IDs to pass to the cross encoder model."""

47 multi_modal_data: NotRequired["MultiModalDataDict"]

@@ -50,7 +50,7 @@ class TokensPrompt(TypedDict):

50 if the model supports it.

51 """

53 - mm_processor_kwargs: NotRequired[Dict[str, Any]]

53 + mm_processor_kwargs: NotRequired[dict[str, Any]]

54 """

55 Optional multi-modal processor kwargs to be forwarded to the

56 multimodal input mapper & processor. Note that if multiple modalities

@@ -115,7 +115,7 @@ class ExplicitEncoderDecoderPrompt(TypedDict, Generic[_T1_co, _T2_co]):

115

116 decoder_prompt: Optional[_T2_co]

117

118 - mm_processor_kwargs: NotRequired[Dict[str, Any]]

118 + mm_processor_kwargs: NotRequired[dict[str, Any]]

119

120

121 PromptType = Union[SingletonPrompt, ExplicitEncoderDecoderPrompt]

@@ -136,10 +136,10 @@ class TokenInputs(TypedDict):

136 type: Literal["token"]

137 """The type of inputs."""

138

139 - prompt_token_ids: List[int]

139 + prompt_token_ids: list[int]

140 """The token IDs of the prompt."""

141

142 - token_type_ids: NotRequired[List[int]]

142 + token_type_ids: NotRequired[list[int]]

143 """The token type IDs of the prompt."""

144

145 prompt: NotRequired[str]

@@ -164,12 +164,12 @@ class TokenInputs(TypedDict):

164 Placeholder ranges for the multi-modal data.

165 """

166

167 - multi_modal_hashes: NotRequired[List[str]]

167 + multi_modal_hashes: NotRequired[list[str]]

168 """

169 The hashes of the multi-modal data.

170 """

171

172 - mm_processor_kwargs: NotRequired[Dict[str, Any]]

172 + mm_processor_kwargs: NotRequired[dict[str, Any]]

173 """

174 Optional multi-modal processor kwargs to be forwarded to the

175 multimodal input mapper & processor. Note that if multiple modalities

@@ -179,14 +179,14 @@ class TokenInputs(TypedDict):

179

180

181 def token_inputs(

182 - prompt_token_ids: List[int],

183 - token_type_ids: Optional[List[int]] = None,

182 + prompt_token_ids: list[int],

183 + token_type_ids: Optional[list[int]] = None,

184 prompt: Optional[str] = None,

185 multi_modal_data: Optional["MultiModalDataDict"] = None,

186 multi_modal_inputs: Optional["MultiModalKwargs"] = None,

187 - multi_modal_hashes: Optional[List[str]] = None,

187 + multi_modal_hashes: Optional[list[str]] = None,

188 multi_modal_placeholders: Optional["MultiModalPlaceholderDict"] = None,

189 - mm_processor_kwargs: Optional[Dict[str, Any]] = None,

189 + mm_processor_kwargs: Optional[dict[str, Any]] = None,

190 ) -> TokenInputs:

191 """Construct :class:`TokenInputs` from optional values."""

192 inputs = TokenInputs(type="token", prompt_token_ids=prompt_token_ids)

@@ -255,7 +255,7 @@ class SingletonInputsAdapter:

255 assert_never(inputs) # type: ignore[arg-type]

256

257 @cached_property

258 - def prompt_token_ids(self) -> List[int]:

258 + def prompt_token_ids(self) -> list[int]:

259 inputs = self.inputs

260

261 if inputs["type"] == "token" or inputs["type"] == "multimodal":

@@ -264,7 +264,7 @@ class SingletonInputsAdapter:

264 assert_never(inputs) # type: ignore[arg-type]

265

266 @cached_property

267 - def token_type_ids(self) -> List[int]:

267 + def token_type_ids(self) -> list[int]:

268 inputs = self.inputs

269

270 if inputs["type"] == "token" or inputs["type"] == "multimodal":

@@ -294,7 +294,7 @@ class SingletonInputsAdapter:

294 assert_never(inputs) # type: ignore[arg-type]

295

296 @cached_property

297 - def multi_modal_inputs(self) -> Union[Dict, "MultiModalKwargs"]:

297 + def multi_modal_inputs(self) -> Union[dict, "MultiModalKwargs"]:

298 inputs = self.inputs

299

300 if inputs["type"] == "token":

@@ -306,7 +306,7 @@ class SingletonInputsAdapter:

306 assert_never(inputs) # type: ignore[arg-type]

307

308 @cached_property

309 - def multi_modal_hashes(self) -> List[str]:

309 + def multi_modal_hashes(self) -> list[str]:

310 inputs = self.inputs

311

312 if inputs["type"] == "token":

@@ -331,7 +331,7 @@ class SingletonInputsAdapter:

331 assert_never(inputs) # type: ignore[arg-type]

332

333 @cached_property

334 - def mm_processor_kwargs(self) -> Dict[str, Any]:

334 + def mm_processor_kwargs(self) -> dict[str, Any]:

335 inputs = self.inputs

336

337 if inputs["type"] == "token":

@@ -355,7 +355,7 @@ _T2 = TypeVar("_T2", bound=SingletonPrompt, default=SingletonPrompt)

355 def build_explicit_enc_dec_prompt(

356 encoder_prompt: _T1,

357 decoder_prompt: Optional[_T2],

358 - mm_processor_kwargs: Optional[Dict[str, Any]] = None,

358 + mm_processor_kwargs: Optional[dict[str, Any]] = None,

359 ) -> ExplicitEncoderDecoderPrompt[_T1, _T2]:

360 if mm_processor_kwargs is None:

361 mm_processor_kwargs = {}

@@ -368,9 +368,9 @@ def build_explicit_enc_dec_prompt(

368 def zip_enc_dec_prompts(

369 enc_prompts: Iterable[_T1],

370 dec_prompts: Iterable[Optional[_T2]],

371 - mm_processor_kwargs: Optional[Union[Iterable[Dict[str, Any]],

372 - Dict[str, Any]]] = None,

373 -) -> List[ExplicitEncoderDecoderPrompt[_T1, _T2]]:

371 + mm_processor_kwargs: Optional[Union[Iterable[dict[str, Any]],

372 + dict[str, Any]]] = None,

373 +) -> list[ExplicitEncoderDecoderPrompt[_T1, _T2]]:

374 """

375 Zip encoder and decoder prompts together into a list of

376 :class:`ExplicitEncoderDecoderPrompt` instances.

@@ -380,12 +380,12 @@ def zip_enc_dec_prompts(

380 provided, it will be zipped with the encoder/decoder prompts.

381 """

382 if mm_processor_kwargs is None:

383 - mm_processor_kwargs = cast(Dict[str, Any], {})

383 + mm_processor_kwargs = cast(dict[str, Any], {})

384 if isinstance(mm_processor_kwargs, dict):

385 return [

386 build_explicit_enc_dec_prompt(

387 encoder_prompt, decoder_prompt,

388 - cast(Dict[str, Any], mm_processor_kwargs))

388 + cast(dict[str, Any], mm_processor_kwargs))

389 for (encoder_prompt,

390 decoder_prompt) in zip(enc_prompts, dec_prompts)

391 ]

@@ -399,7 +399,7 @@ def zip_enc_dec_prompts(

399

400 def to_enc_dec_tuple_list(

401 enc_dec_prompts: Iterable[ExplicitEncoderDecoderPrompt[_T1, _T2]],

402 -) -> List[Tuple[_T1, Optional[_T2]]]:

402 +) -> list[tuple[_T1, Optional[_T2]]]:

403 return [(enc_dec_prompt["encoder_prompt"],

404 enc_dec_prompt["decoder_prompt"])

405 for enc_dec_prompt in enc_dec_prompts]

vllm/inputs/parse.py

+9 -8

@@ -1,6 +1,7 @@

1 # SPDX-License-Identifier: Apache-2.0

3 -from typing import List, Literal, Sequence, TypedDict, Union, cast, overload

3 +from collections.abc import Sequence

4 +from typing import Literal, TypedDict, Union, cast, overload

6 from typing_extensions import TypeIs

@@ -17,24 +18,24 @@ class ParsedText(TypedDict):

20 class ParsedTokens(TypedDict):

20 - content: List[int]

21 + content: list[int]

22 is_tokens: Literal[True]

25 @overload

26 def parse_and_batch_prompt(

26 - prompt: Union[str, List[str]]) -> Sequence[ParsedText]:

27 + prompt: Union[str, list[str]]) -> Sequence[ParsedText]:

28 ...

31 @overload

32 def parse_and_batch_prompt(

32 - prompt: Union[List[int], List[List[int]]]) -> Sequence[ParsedTokens]:

33 + prompt: Union[list[int], list[list[int]]]) -> Sequence[ParsedTokens]:

34 ...

37 def parse_and_batch_prompt(

37 - prompt: Union[str, List[str], List[int], List[List[int]]],

38 + prompt: Union[str, list[str], list[int], list[list[int]]],

39 ) -> Union[Sequence[ParsedText], Sequence[ParsedTokens]]:

40 if isinstance(prompt, str):

41 # case 1: a string

@@ -46,16 +47,16 @@ def parse_and_batch_prompt(

48 if is_list_of(prompt, str):

49 # case 2: array of strings

49 - prompt = cast(List[str], prompt)

50 + prompt = cast(list[str], prompt)

51 return [

52 ParsedText(content=elem, is_tokens=False) for elem in prompt

53 ]

54 if is_list_of(prompt, int):

55 # case 3: array of tokens

55 - prompt = cast(List[int], prompt)

56 + prompt = cast(list[int], prompt)

57 return [ParsedTokens(content=prompt, is_tokens=True)]

58 if is_list_of(prompt, list):

58 - prompt = cast(List[List[int]], prompt)

59 + prompt = cast(list[list[int]], prompt)

60 if len(prompt[0]) == 0:

61 raise ValueError("please provide at least one prompt")

vllm/inputs/preprocess.py

+12 -11

@@ -1,7 +1,8 @@

1 # SPDX-License-Identifier: Apache-2.0

3 import asyncio

4 -from typing import List, Mapping, Optional, Tuple, Union, cast

4 +from collections.abc import Mapping

5 +from typing import Optional, Union, cast

7 from typing_extensions import assert_never

@@ -92,7 +93,7 @@ class InputPreprocessor:

94 return dec_start_token_id

95 - def _get_default_enc_dec_decoder_prompt(self) -> List[int]:

96 + def _get_default_enc_dec_decoder_prompt(self) -> list[int]:

97 '''

98 Specifically for encoder/decoder models:

99 generate a default decoder prompt for when

@@ -130,8 +131,8 @@ class InputPreprocessor:

131

132 def _prepare_decoder_input_ids_for_generation(

133 self,

133 - decoder_input_ids: Optional[List[int]],

134 - ) -> List[int]:

134 + decoder_input_ids: Optional[list[int]],

135 + ) -> list[int]:

136 """

137 Prepares `decoder_input_ids` for generation with encoder-decoder models.

138

@@ -168,9 +169,9 @@ class InputPreprocessor:

169

170 def _apply_prompt_adapter(

171 self,

171 - prompt_token_ids: List[int],

172 + prompt_token_ids: list[int],

173 prompt_adapter_request: Optional[PromptAdapterRequest],

173 - ) -> List[int]:

174 + ) -> list[int]:

175 if prompt_adapter_request:

176 prompt_token_ids = (

177 [0] * prompt_adapter_request.prompt_adapter_num_virtual_tokens

@@ -183,7 +184,7 @@ class InputPreprocessor:

184 prompt: str,

185 request_id: str,

186 lora_request: Optional[LoRARequest],

186 - ) -> List[int]:

187 + ) -> list[int]:

188 """

189 Apply the model's tokenizer to a text prompt, returning the

190 corresponding token IDs.

@@ -211,7 +212,7 @@ class InputPreprocessor:

212 prompt: str,

213 request_id: str,

214 lora_request: Optional[LoRARequest],

214 - ) -> List[int]:

215 + ) -> list[int]:

216 """Async version of :meth:`_tokenize_prompt`."""

217 tokenizer = self.get_tokenizer_group()

218 add_special_tokens = None

@@ -250,7 +251,7 @@ class InputPreprocessor:

251

252 def _process_multimodal(

253 self,

253 - prompt: Union[str, List[int]],

254 + prompt: Union[str, list[int]],

255 mm_data: MultiModalDataDict,

256 mm_processor_kwargs: Optional[Mapping[str, object]],

257 lora_request: Optional[LoRARequest],

@@ -280,7 +281,7 @@ class InputPreprocessor:

281

282 async def _process_multimodal_async(

283 self,

283 - prompt: Union[str, List[int]],

284 + prompt: Union[str, list[int]],

285 mm_data: MultiModalDataDict,

286 mm_processor_kwargs: Optional[Mapping[str, object]],

287 lora_request: Optional[LoRARequest],

@@ -511,7 +512,7 @@ class InputPreprocessor:

512 self,

513 inputs: SingletonInputs,

514 decoder_inputs_to_override: Optional[SingletonInputs] = None,

514 - ) -> Tuple[SingletonInputs, SingletonInputs]:

515 + ) -> tuple[SingletonInputs, SingletonInputs]:

516 """

517 For encoder/decoder models only:

518 Separate Encoder/Decoder inputs from a MultiModalEncDecInputs

vllm/inputs/registry.py

+3 -2

@@ -2,9 +2,10 @@

3 import functools

4 from collections import UserDict

5 +from collections.abc import Mapping

6 from dataclasses import dataclass

6 -from typing import (TYPE_CHECKING, Any, Callable, Mapping, NamedTuple,

7 - Optional, Protocol, Union)

7 +from typing import (TYPE_CHECKING, Any, Callable, NamedTuple, Optional,

8 + Protocol, Union)

10 from torch import nn

11 from transformers import BatchFeature, PretrainedConfig, ProcessorMixin

vllm/multimodal/base.py

+4 -3

@@ -2,9 +2,10 @@

3 from abc import ABC, abstractmethod

4 from collections import defaultdict

5 +from collections.abc import Sequence

6 from pathlib import Path

7 from typing import (TYPE_CHECKING, Any, Callable, Generic, NamedTuple,

7 - Optional, Sequence, Tuple, Type, TypeVar, Union)

8 + Optional, TypeVar, Union)

10 from torch import nn

@@ -39,7 +40,7 @@ model. This does not include tokens that correspond to the input text.

40 """

42 _T = TypeVar("_T")

42 -N = TypeVar("N", bound=Type[nn.Module])

43 +N = TypeVar("N", bound=type[nn.Module])

46 class MultiModalPlugin(ABC):

@@ -274,7 +275,7 @@ class MultiModalPlaceholderMap:

275 @classmethod

276 def from_seq_group(

277 cls, seq_group: "SequenceGroupMetadata", positions: range

277 - ) -> Tuple[Optional[MultiModalDataDict], dict[str,

278 + ) -> tuple[Optional[MultiModalDataDict], dict[str,

279 "MultiModalPlaceholderMap"]]:

280 """

281 Returns the multi-modal items that intersect with the portion of a

vllm/multimodal/hasher.py

+2 -1

@@ -1,7 +1,8 @@

1 # SPDX-License-Identifier: Apache-2.0

3 import pickle

4 -from typing import TYPE_CHECKING, Iterable, Mapping, Optional

4 +from collections.abc import Iterable, Mapping

5 +from typing import TYPE_CHECKING, Optional

7 import numpy as np

8 import torch

vllm/multimodal/image.py

+2 -2

@@ -3,7 +3,7 @@

3 import base64

4 from io import BytesIO

5 from pathlib import Path

6 -from typing import TYPE_CHECKING, Any, Dict, Optional

6 +from typing import TYPE_CHECKING, Any, Optional

8 import torch

9 from PIL import Image

@@ -31,7 +31,7 @@ class ImagePlugin(MultiModalPlugin):

31 def _get_hf_image_processor(

32 self,

33 model_config: "ModelConfig",

34 - mm_processor_kwargs: Optional[Dict[str, Any]] = None,

34 + mm_processor_kwargs: Optional[dict[str, Any]] = None,

35 ):

36 if mm_processor_kwargs is None:

37 mm_processor_kwargs = {}

vllm/multimodal/registry.py

+7 -7

@@ -2,9 +2,9 @@

3 import functools

4 from collections import UserDict

5 +from collections.abc import Mapping, Sequence

6 from dataclasses import dataclass

6 -from typing import (TYPE_CHECKING, Any, Dict, Generic, Mapping, Optional,

7 - Protocol, Sequence, Type, TypeVar)

7 +from typing import TYPE_CHECKING, Any, Generic, Optional, Protocol, TypeVar

9 import torch.nn as nn

@@ -29,7 +29,7 @@ if TYPE_CHECKING:

30 logger = init_logger(__name__)

32 -N = TypeVar("N", bound=Type[nn.Module])

32 +N = TypeVar("N", bound=type[nn.Module])

33 _I = TypeVar("_I", bound=BaseProcessingInfo)

34 _I_co = TypeVar("_I_co", bound=BaseProcessingInfo, covariant=True)

@@ -83,13 +83,13 @@ class _ProcessorFactories(Generic[_I]):

83 return self.processor(info, dummy_inputs_builder, cache=cache)

86 -class _MultiModalLimits(UserDict["ModelConfig", Dict[str, int]]):

86 +class _MultiModalLimits(UserDict["ModelConfig", dict[str, int]]):

87 """

88 Wraps `_limits_by_model` for a more informative error message

89 when attempting to access a model that does not exist.

90 """

92 - def __getitem__(self, key: "ModelConfig") -> Dict[str, int]:

92 + def __getitem__(self, key: "ModelConfig") -> dict[str, int]:

93 try:

94 return super().__getitem__(key)

95 except KeyError as exc:

@@ -170,7 +170,7 @@ class MultiModalRegistry:

170 self,

171 model_config: "ModelConfig",

172 data: MultiModalDataDict,

173 - mm_processor_kwargs: Optional[Dict[str, Any]] = None,

173 + mm_processor_kwargs: Optional[dict[str, Any]] = None,

174 ) -> MultiModalKwargs:

175 """

176 Apply an input mapper to the data passed to the model.

@@ -184,7 +184,7 @@ class MultiModalRegistry:

184 Note:

185 This should be called after :meth:`init_mm_limits_per_prompt`.

186 """

187 - merged_dict: Dict[str, NestedTensors] = {}

187 + merged_dict = dict[str, NestedTensors]()

188

189 for data_key, data_value in data.items():

190 plugin = self._get_plugin(data_key)

vllm/multimodal/video.py

+2 -2

@@ -4,7 +4,7 @@ import base64

4 from functools import partial

5 from io import BytesIO

6 from pathlib import Path

7 -from typing import TYPE_CHECKING, Any, Dict, Optional

7 +from typing import TYPE_CHECKING, Any, Optional

9 import numpy as np

10 import numpy.typing as npt

@@ -39,7 +39,7 @@ class VideoPlugin(ImagePlugin):

39 def _get_hf_video_processor(

40 self,

41 model_config: "ModelConfig",

42 - mm_processor_kwargs: Optional[Dict[str, Any]] = None,

42 + mm_processor_kwargs: Optional[dict[str, Any]] = None,

43 ):

44 if mm_processor_kwargs is None:

45 mm_processor_kwargs = {}

vllm/usage/usage_lib.py

+7 -7

@@ -9,7 +9,7 @@ import time

9 from enum import Enum

10 from pathlib import Path

11 from threading import Thread

12 -from typing import Any, Dict, Optional, Union

12 +from typing import Any, Optional, Union

13 from uuid import uuid4

15 import cpuinfo

@@ -27,7 +27,7 @@ _USAGE_STATS_DO_NOT_TRACK_PATH = os.path.join(_config_home, "do_not_track")

27 _USAGE_STATS_ENABLED = None

28 _USAGE_STATS_SERVER = envs.VLLM_USAGE_STATS_SERVER

30 -_GLOBAL_RUNTIME_DATA: Dict[str, Union[str, int, bool]] = {}

30 +_GLOBAL_RUNTIME_DATA = dict[str, Union[str, int, bool]]()

32 _USAGE_ENV_VARS_TO_COLLECT = [

33 "VLLM_USE_MODELSCOPE",

@@ -150,7 +150,7 @@ class UsageMessage:

150 def report_usage(self,

151 model_architecture: str,

152 usage_context: UsageContext,

153 - extra_kvs: Optional[Dict[str, Any]] = None) -> None:

153 + extra_kvs: Optional[dict[str, Any]] = None) -> None:

154 t = Thread(target=self._report_usage_worker,

155 args=(model_architecture, usage_context, extra_kvs or {}),

156 daemon=True)

@@ -158,13 +158,13 @@ class UsageMessage:

158

159 def _report_usage_worker(self, model_architecture: str,

160 usage_context: UsageContext,

161 - extra_kvs: Dict[str, Any]) -> None:

161 + extra_kvs: dict[str, Any]) -> None:

162 self._report_usage_once(model_architecture, usage_context, extra_kvs)

163 self._report_continous_usage()

164

165 def _report_usage_once(self, model_architecture: str,

166 usage_context: UsageContext,

167 - extra_kvs: Dict[str, Any]) -> None:

167 + extra_kvs: dict[str, Any]) -> None:

168 # Platform information

169 from vllm.platforms import current_platform

170 if current_platform.is_cuda_alike():

@@ -227,7 +227,7 @@ class UsageMessage:

227 self._write_to_file(data)

228 self._send_to_server(data)

229

230 - def _send_to_server(self, data: Dict[str, Any]) -> None:

230 + def _send_to_server(self, data: dict[str, Any]) -> None:

231 try:

232 global_http_client = global_http_connection.get_sync_client()

233 global_http_client.post(_USAGE_STATS_SERVER, json=data)

@@ -235,7 +235,7 @@ class UsageMessage:

235 # silently ignore unless we are using debug log

236 logging.debug("Failed to send usage data to server")

237

238 - def _write_to_file(self, data: Dict[str, Any]) -> None:

238 + def _write_to_file(self, data: dict[str, Any]) -> None:

239 os.makedirs(os.path.dirname(_USAGE_STATS_JSON_PATH), exist_ok=True)

240 Path(_USAGE_STATS_JSON_PATH).touch(exist_ok=True)

241 with open(_USAGE_STATS_JSON_PATH, "a") as f: