Add pydocstyle to the ruff rules (#52136)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Harry Mellor <hej.mellor@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-16 16:44:29 +00:00
committed by GitHub
co-authored by Claude Opus 5
parent 804e5377aa
commit 8c1557a79c
2002 changed files with 15024 additions and 14269 deletions
@@ -224,7 +224,8 @@ class _MultiModalProcessorBase(BaseMultiModalProcessor[MultiModalProcessingInfo]
mm_items: MultiModalDataItems,
) -> tuple[Mapping[str, object], Mapping[str, object]]:
"""Rename the parser's `audios` key to the `audio` argument HF audio
processors take."""
processors take.
"""
processor_data, passthrough_data = super()._get_hf_mm_data(mm_items)
if self.info._is_audio_model() and "audios" in processor_data:
processor_data["audio"] = processor_data.pop("audios")
@@ -432,7 +433,8 @@ class LegacyMultiModalProcessor(_MultiModalProcessorBase):
out_mm_kwargs: MultiModalKwargsItems,
) -> Sequence[PromptUpdate]:
"""Empty, because `apply` writes the placeholder ranges itself rather than
deriving them from updates."""
deriving them from updates.
"""
return []
def _get_mm_token_ids(self, modality: str) -> list[int]:
@@ -466,7 +468,8 @@ class LegacyMultiModalProcessor(_MultiModalProcessorBase):
num_audios: int,
) -> dict[str, list[PlaceholderRange]]:
"""Take each contiguous run of the audio token as one item's placeholder,
and record how many tokens the run holds."""
and record how many tokens the run holds.
"""
audio_token_ids = self._get_mm_token_ids("audio")
prompt_tensor = torch.tensor(prompt_ids)
is_audio = torch.isin(prompt_tensor, torch.tensor(audio_token_ids))
@@ -524,7 +527,8 @@ class LegacyMultiModalProcessor(_MultiModalProcessorBase):
mm_token_type_ids: torch.Tensor | None,
) -> dict[str, list[PlaceholderRange]]:
"""Split the positions the processor marks as image into one placeholder per
item, sized by the token count it reports for each image."""
item, sized by the token count it reports for each image.
"""
hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs)
if mm_token_type_ids is None:
raise ValueError(
@@ -577,7 +581,8 @@ class LegacyMultiModalProcessor(_MultiModalProcessorBase):
timing_ctx: TimingContext,
) -> MultiModalInput:
"""Process the prompt and every multi-modal item in one HF processor call,
then read the placeholder ranges out of the token ids it returns."""
then read the placeholder ranges out of the token ids it returns.
"""
prompt = inputs.prompt
mm_items = inputs.mm_data_items
hf_processor_mm_kwargs = inputs.hf_processor_mm_kwargs
@@ -705,7 +710,8 @@ class OffsetsMultiModalProcessor(_MultiModalProcessorBase):
out_mm_kwargs: MultiModalKwargsItems,
) -> Sequence[PromptUpdate]:
"""Replace each modality's placeholder token with the token ids that item's
replacement text encodes to, marking which of them hold embeddings."""
replacement text encodes to, marking which of them hold embeddings.
"""
hf_processor = self.info.get_hf_processor(**hf_processor_mm_kwargs)
tokenizer = self.info.get_tokenizer()
@@ -892,7 +898,8 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
self, model: "PreTrainedModel"
) -> dict[str, type["PreTrainedModel"]]:
"""Modalities whose encoder cannot be told apart from the model itself are
omitted, as are those `get_encoder` rejects."""
omitted, as are those `get_encoder` rejects.
"""
encoder_classes: dict[str, type[PreTrainedModel]] = {}
for modality in _MODALITY_TO_TOKEN_TYPE_ID:
try:
@@ -953,9 +960,8 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
yield
def _decorate_for_torch_compile(self):
"""
Decorate the model's decoder and encoder classes to indicate to vLLM that they
support torch compile if `can_enable_torch_compile` and
"""Decorate the model's decoder and encoder classes to indicate to vLLM
that they support torch compile if `can_enable_torch_compile` and
`should_torch_compile_mm_encoder` are True respectively.
"""
super()._decorate_for_torch_compile()
@@ -1012,8 +1018,8 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
def get_language_model(self) -> torch.nn.Module:
"""Transformers modeling backend multimodal classes do not contain a separate
vLLM language model class. Therefore, in order to return a language model vLLM
class, we use a wrapper to give `self` the same interface as a text model."""
class, we use a wrapper to give `self` the same interface as a text model.
"""
# Exclude self and object
bases = self.__class__.mro()[1:-1]
# Keep only classes defined in `vllm.model_executor.models.transformers`
@@ -1031,9 +1037,7 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
return LanguageModel(self)
def get_mm_mapping(self) -> MultiModelKeys:
"""
Get the module prefix in multimodal models
"""
"""Get the module prefix in multimodal models"""
for name in ("language_model", "text_model"):
if getattr(self.model, name, None) is not None:
return MultiModelKeys.from_string_field(language_model=f"model.{name}")
@@ -1169,7 +1173,8 @@ class MultiModalMixin(SupportsMultiModal, SupportsMRoPE):
) -> dict[str, Any]:
"""Narrow the entries of `kwargs` that hold one row per item down to the item
at `index`. Length is all there is to match on, so an unrelated entry of the
same length is narrowed too."""
same length is narrowed too.
"""
return {
key: value[index : index + 1]
if isinstance(value, (torch.Tensor, list)) and len(value) == num_items