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
+17 -15
View File
@@ -3,7 +3,7 @@
from abc import abstractmethod
from collections.abc import Iterable, Mapping
from typing import Annotated, Final, Literal, Optional, Protocol, TypeVar, Union
from typing import Annotated, Final, Literal, Protocol, TypeAlias, TypeVar
import torch
import torch.nn as nn
@@ -55,11 +55,11 @@ class LlavaNextImagePixelInputs(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)]
# This should be in `(height, width)` format.
@@ -75,7 +75,9 @@ class LlavaNextImageEmbeddingInputs(TensorSchema):
data: Annotated[torch.Tensor, TensorShape("bn", "ifs", "hs")]
LlavaNextImageInputs = Union[LlavaNextImagePixelInputs, LlavaNextImageEmbeddingInputs]
LlavaNextImageInputs: TypeAlias = (
LlavaNextImagePixelInputs | LlavaNextImageEmbeddingInputs
)
class LlavaNextLikeConfig(LlavaLikeConfig, Protocol):
@@ -235,7 +237,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
)
@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>"
@@ -294,7 +296,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
def _parse_and_validate_image_input(
self, **kwargs: object
) -> Optional[LlavaNextImageInputs]:
) -> LlavaNextImageInputs | None:
pixel_values = kwargs.pop("pixel_values", None)
image_sizes = kwargs.pop("image_sizes", None)
image_embeds = kwargs.pop("image_embeds", None)
@@ -324,7 +326,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
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
@@ -424,7 +426,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
def _process_image_pixels(
self,
inputs: LlavaNextImagePixelInputs,
) -> Union[torch.Tensor, tuple[torch.Tensor, ...]]:
) -> torch.Tensor | tuple[torch.Tensor, ...]:
assert self.vision_tower is not None
pixel_values = inputs["pixel_values"]
@@ -456,7 +458,7 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
def _process_image_input(
self,
image_input: LlavaNextImageInputs,
) -> Union[torch.Tensor, list[torch.Tensor]]:
) -> torch.Tensor | list[torch.Tensor]:
if image_input["type"] == "image_embeds":
return [image_input["data"]]
@@ -491,9 +493,9 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
def get_input_embeddings(
self,
input_ids: torch.Tensor,
multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
multimodal_embeddings: MultiModalEmbeddings | None = None,
*,
is_multimodal: Optional[torch.Tensor] = None,
is_multimodal: torch.Tensor | None = None,
# Multi-modal token ID may exceed vocab size
handle_oov_mm_token: bool = True,
) -> torch.Tensor:
@@ -512,10 +514,10 @@ class LlavaNextForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsP
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-NeXT.
One key thing to understand is the `input_ids` already accounts for the
@@ -573,7 +575,7 @@ model_executor.models.llava_next.LlavaNextProcessingInfo.get_num_image_tokens].
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]: