@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 22cf679aad

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 14, 05:30 UTC

[Doc]: fix various typos in multiple files (#23179)

Signed-off-by: Didier Durand <durand.didier@gmail.com>

Didier Durand committed Aug 22, 2025 at 19:38 UTC 22cf679aadca99311cfb5a9f894039e464e366aa

7 files changed +12 -12

vllm/beam_search.py

+1 -1

@@ -18,7 +18,7 @@ class BeamSearchSequence:

18 The text field is optional and will only be filled when the sequence is

19 about to be returned to the user.

20 """

21 - # The tokens includes the prompt.

21 + # The tokens include the prompt.

22 tokens: list[int]

23 logprobs: list[dict[int, Logprob]]

24 lora_request: Optional[LoRARequest] = None

vllm/compilation/backends.py

+1 -1

@@ -484,7 +484,7 @@ class VllmBackend:

484

485 factors = []

486 # 0. factors come from the env, for example, The values of

487 - # VLLM_PP_LAYER_PARTITION will affects the computation graph.

487 + # VLLM_PP_LAYER_PARTITION will affect the computation graph.

488 env_hash = envs.compute_hash()

489 factors.append(env_hash)

490

vllm/engine/arg_utils.py

+3 -3

@@ -605,7 +605,7 @@ class EngineArgs:

605 **guided_decoding_kwargs["disable_additional_properties"])

606 guided_decoding_group.add_argument(

607 "--reasoning-parser",

608 - # This choices is a special case because it's not static

608 + # This choice is a special case because it's not static

609 choices=list(ReasoningParserManager.reasoning_parsers),

610 **guided_decoding_kwargs["reasoning_backend"])

611

@@ -1047,7 +1047,7 @@ class EngineArgs:

1047 # details from the config directly

1048 # no user input required / expected

1049 if isinstance(hf_config, SpeculatorsConfig):

1050 - # We create one since we dont create one

1050 + # We create one since we don't create one

1051 self.speculative_config = {}

1052 self.speculative_config[

1053 "num_speculative_tokens"] = hf_config.num_lookahead_tokens

@@ -1775,7 +1775,7 @@ class AsyncEngineArgs(EngineArgs):

1775 def add_cli_args(parser: FlexibleArgumentParser,

1776 async_args_only: bool = False) -> FlexibleArgumentParser:

1777 # Initialize plugin to update the parser, for example, The plugin may

1778 - # adding a new kind of quantization method to --quantization argument or

1778 + # add a new kind of quantization method to --quantization argument or

1779 # a new device to --device argument.

1780 load_general_plugins()

1781 if not async_args_only:

vllm/engine/multiprocessing/client.py

+2 -2

@@ -539,7 +539,7 @@ class MQLLMEngineClient(EngineClient):

539 if request_id in self.output_queues:

540 raise ValueError(f"Request {request_id} already exists")

541

542 - # 1) Create output queue for this requests.

542 + # 1) Create output queue for this request.

543 queue: asyncio.Queue[Union[RequestOutput,

544 BaseException]] = asyncio.Queue()

545 self.output_queues[request_id] = queue

@@ -651,7 +651,7 @@ class MQLLMEngineClient(EngineClient):

651 # Uses the same I/O as generate requests

652 request = RPCLoadAdapterRequest(lora_request)

653

654 - # Create output queue for this requests.

654 + # Create output queue for this request.

655 queue: asyncio.Queue[Union[None, BaseException]] = asyncio.Queue()

656 self.output_queues[request.request_id] = queue

657

vllm/entrypoints/chat_utils.py

+1 -1

@@ -1330,7 +1330,7 @@ def apply_mistral_chat_template(

1330 # mistral-common uses assert statements to stop processing of input

1331 # if input does not comply with the expected format.

1332 # We convert those assertion errors to ValueErrors so they can be

1333 - # are properly caught in the preprocessing_input step

1333 + # properly caught in the preprocessing_input step

1334 except (AssertionError, MistralCommonException) as e:

1335 raise ValueError(str(e)) from e

1336

vllm/utils/__init__.py

+2 -2

@@ -2482,7 +2482,7 @@ class PlaceholderModule(_PlaceholderBase):

2482 A placeholder object to use when a module does not exist.

2483

2484 This enables more informative errors when trying to access attributes

2485 - of a module that does not exists.

2485 + of a module that does not exist.

2486 """

2487

2488 def __init__(self, name: str) -> None:

@@ -3109,7 +3109,7 @@ class LazyLoader(types.ModuleType):

3109 """

3110 LazyLoader module borrowed from Tensorflow

3111 https://github.com/tensorflow/tensorflow/blob/main/tensorflow/python/util/lazy_loader.py

3112 - with a addition of "module caching".

3112 + with an addition of "module caching".

3113

3114 Lazily import a module, mainly to avoid pulling in large dependencies.

3115 Modules such as `xgrammar` might do additional side effects, so we

vllm/v1/structured_output/__init__.py

+2 -2

@@ -267,7 +267,7 @@ class StructuredOutputManager:

267 assert request.structured_output_request is not None

268 assert request.structured_output_request.grammar is not None

269 # by default, we should always advance

270 - # for cases that doesn't uses thinking mode.

270 + # for cases that don't use thinking mode.

271 if self.reasoner is not None:

272 structured_req = request.structured_output_request

273

@@ -276,7 +276,7 @@ class StructuredOutputManager:

276

277 # Check if reasoning ends in *this* step

278 if self.reasoner.is_reasoning_end(request.all_token_ids):

279 - # Reasoning just ended, so we shouldn't advanced til

279 + # Reasoning just ended, so we shouldn't advance til

280 # next pass

281 structured_req.reasoning_ended = True

282