Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-12 09:51:31 -07:00
committed by GitHub
parent 9bb38130cb
commit 8fcaaf6a16
944 changed files with 9490 additions and 10121 deletions
+23 -23
View File
@@ -3,7 +3,7 @@
import math
from collections.abc import Iterable, Mapping, Sequence
from typing import Annotated, Final, Literal, Optional, Protocol, Union
from typing import Annotated, Final, Literal, Protocol, TypeAlias
import torch
import torch.nn as nn
@@ -69,7 +69,7 @@ class LlavaOnevisionVideoPixelInputs(TensorSchema):
type: Literal["pixel_values_videos"] = "pixel_values_videos"
pixel_values_videos: Annotated[
Union[torch.Tensor, list[torch.Tensor]],
torch.Tensor | list[torch.Tensor],
TensorShape("bn", "f", 3, "h", "w", dynamic_dims={"f"}),
]
@@ -90,11 +90,11 @@ class LlavaOnevisionImagePixelInputs(TensorSchema):
type: Literal["pixel_values"] = "pixel_values"
pixel_values: Annotated[
Union[torch.Tensor, list[torch.Tensor]],
torch.Tensor | list[torch.Tensor],
TensorShape("bn", "np", 3, "h", "w", dynamic_dims={"np"}),
]
image_sizes: Annotated[Optional[torch.Tensor], TensorShape("bn", 2)]
image_sizes: Annotated[torch.Tensor | None, TensorShape("bn", 2)]
class LlavaOnevisionImageEmbeddingInputs(TensorSchema):
@@ -113,13 +113,13 @@ class LlavaOnevisionImageEmbeddingInputs(TensorSchema):
]
LlavaOnevisionImageInputs = Union[
LlavaOnevisionImagePixelInputs, LlavaOnevisionImageEmbeddingInputs
]
LlavaOnevisionImageInputs: TypeAlias = (
LlavaOnevisionImagePixelInputs | LlavaOnevisionImageEmbeddingInputs
)
LlavaOnevisionMultiInputs = Union[
LlavaOnevisionImageInputs, LlavaOnevisionVideoPixelInputs
]
LlavaOnevisionMultiInputs: TypeAlias = (
LlavaOnevisionImageInputs | LlavaOnevisionVideoPixelInputs
)
class LlavaOnevisionLikeConfig(LlavaNextLikeConfig, Protocol):
@@ -133,7 +133,7 @@ class LlavaOnevisionProcessingInfo(LlavaNextProcessingInfo):
def get_hf_processor(self, **kwargs: object):
return self.ctx.get_hf_processor(LlavaOnevisionProcessor, **kwargs)
def get_supported_mm_limits(self) -> Mapping[str, Optional[int]]:
def get_supported_mm_limits(self) -> Mapping[str, int | None]:
return {"image": None, "video": None}
# Based on: https://github.com/huggingface/text-generation-inference/blob/v3.0.1/server/text_generation_server/models/vlm_causal_lm.py#L86
@@ -276,7 +276,7 @@ class LlavaOnevisionDummyInputsBuilder(
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
mm_options: Mapping[str, BaseDummyOptions] | None = None,
) -> MultiModalDataDict:
num_images = mm_counts.get("image", 0)
num_videos = mm_counts.get("video", 0)
@@ -493,7 +493,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
)
@classmethod
def get_placeholder_str(cls, modality: str, i: int) -> Optional[str]:
def get_placeholder_str(cls, modality: str, i: int) -> str | None:
if modality.startswith("image"):
return "<image>"
if modality.startswith("video"):
@@ -531,7 +531,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _parse_and_validate_image_input(
self, **kwargs: object
) -> Optional[LlavaOnevisionImageInputs]:
) -> LlavaOnevisionImageInputs | None:
pixel_values = kwargs.pop("pixel_values", None)
image_sizes = kwargs.pop("image_sizes", None)
image_embeds = kwargs.pop("image_embeds", None)
@@ -560,7 +560,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _parse_and_validate_video_input(
self, **kwargs: object
) -> Optional[LlavaOnevisionVideoPixelInputs]:
) -> LlavaOnevisionVideoPixelInputs | None:
"""
A legal video input should have the following dimensions:
{
@@ -606,7 +606,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _image_pixels_to_features(
self,
vision_tower: Union[CLIPVisionModel, SiglipVisionModel],
vision_tower: CLIPVisionModel | SiglipVisionModel,
pixel_values: torch.Tensor,
) -> torch.Tensor:
# NOTE: we skip the step to select the vision feature layer since
@@ -726,7 +726,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _process_image_pixels(
self,
inputs: LlavaOnevisionImagePixelInputs,
) -> Union[torch.Tensor, list[torch.Tensor]]:
) -> torch.Tensor | list[torch.Tensor]:
assert self.vision_tower is not None
pixel_values = inputs["pixel_values"]
@@ -761,7 +761,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _process_image_input(
self,
image_input: LlavaOnevisionImageInputs,
) -> Union[torch.Tensor, list[torch.Tensor]]:
) -> torch.Tensor | list[torch.Tensor]:
if image_input["type"] == "image_embeds":
return [image_input["data"]]
@@ -788,7 +788,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def _video_pixels_to_features(
self,
vision_tower: Union[CLIPVisionModel, SiglipVisionModel],
vision_tower: CLIPVisionModel | SiglipVisionModel,
pixel_values: torch.Tensor,
) -> torch.Tensor:
# NOTE: we skip the step to select the vision feature layer since
@@ -893,10 +893,10 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
self,
input_ids: torch.Tensor,
positions: torch.Tensor,
intermediate_tensors: Optional[IntermediateTensors] = None,
inputs_embeds: Optional[torch.Tensor] = None,
intermediate_tensors: IntermediateTensors | None = None,
inputs_embeds: torch.Tensor | None = None,
**kwargs: object,
) -> Union[torch.Tensor, IntermediateTensors]:
) -> torch.Tensor | IntermediateTensors:
"""Run forward pass for LlaVA-Onevision.
Args:
input_ids: Flattened (concatenated) input_ids corresponding to a
@@ -915,7 +915,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
def compute_logits(
self,
hidden_states: torch.Tensor,
) -> Optional[torch.Tensor]:
) -> torch.Tensor | None:
return self.language_model.compute_logits(hidden_states)
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]: