transformers 5.0.0rc2__py3-none-any.whl → 5.1.0__py3-none-any.whl

transformers/models/edgetam_video/modular_edgetam_video.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 the HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -15,12 +14,11 @@
 
 import math
 from collections.abc import Callable
-from typing import Any, Optional
+from typing import Any
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-import torch.utils.checkpoint
 from torch import Tensor
 
 from transformers.models.sam2.modeling_sam2 import (
@@ -374,7 +372,7 @@ class EdgeTamVideoVisionEncoderOutput(Sam2VideoVisionEncoderOutput):
 
 
 class EdgeTamVideoVisionRotaryEmbedding(Sam2VideoVisionRotaryEmbedding):
-    def __init__(self, config: EdgeTamVideoConfig, end_x: Optional[int] = None, end_y: Optional[int] = None):
+    def __init__(self, config: EdgeTamVideoConfig, end_x: int | None = None, end_y: int | None = None):
         nn.Module.__init__()
         self.dim = config.memory_attention_hidden_size // (
             config.memory_attention_downsample_rate * config.memory_attention_num_attention_heads
@@ -538,9 +536,9 @@ class EdgeTamVideoRoPESelfAttention(nn.Module):
         # Apply rotary position encoding for self-attention
         query, key = apply_rotary_pos_emb_2d_self_attn(query, key, cos=cos, sin=sin)
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -614,9 +612,9 @@ class EdgeTamVideoRoPECrossAttention(nn.Module):
             num_k_exclude_rope=num_k_exclude_rope,
         )
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -706,7 +704,7 @@ class EdgeTamVideoMemoryAttentionLayer(nn.Module):
         keys: Tensor,
         key_point_embedding: Tensor,
         rope_position_embeddings: tuple[Tensor, Tensor],
-        rope_position_embeddings_k: Optional[tuple[Tensor, Tensor]] = None,
+        rope_position_embeddings_k: tuple[Tensor, Tensor] | None = None,
         num_k_exclude_rope: int = 0,
         rope_k_repeat: int = 0,
     ) -> torch.Tensor:
@@ -745,8 +743,8 @@ class EdgeTamVideoMemoryAttention(Sam2VideoMemoryAttention):
         self,
         current_vision_features: torch.Tensor,
         memory: torch.Tensor,
-        current_vision_position_embeddings: Optional[Tensor] = None,
-        memory_posision_embeddings: Optional[Tensor] = None,
+        current_vision_position_embeddings: Tensor | None = None,
+        memory_posision_embeddings: Tensor | None = None,
         num_object_pointer_tokens: int = 0,
         num_spatial_memory_tokens: int = -1,
     ):
@@ -832,7 +830,7 @@ class EdgeTamVideoPerceiverAttention(nn.Module):
         query: torch.Tensor,
         key: torch.Tensor,
         value: torch.Tensor,
-        positional_encoding: Optional[torch.Tensor] = None,
+        positional_encoding: torch.Tensor | None = None,
         **kwargs,
     ) -> torch.Tensor:
         # Project queries, keys, and values
@@ -856,9 +854,9 @@ class EdgeTamVideoPerceiverAttention(nn.Module):
             value = value + pos_encoding
 
         # Apply attention
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, _ = attention_interface(
             self,
@@ -897,7 +895,7 @@ class EdgeTamVideoPerceiverEncoderLayer(nn.Module):
         self,
         latents: torch.Tensor,
         input_features: torch.Tensor,
-        positional_encoding: Optional[torch.Tensor] = None,
+        positional_encoding: torch.Tensor | None = None,
     ) -> torch.Tensor:
         # Cross attention with layer norms
         normalized_latents = self.layer_norm_latents(latents)
@@ -951,8 +949,8 @@ class EdgeTamVideoPerceiverResampler(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        positional_encoding: Optional[torch.Tensor] = None,
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        positional_encoding: torch.Tensor | None = None,
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
         output_latents = []
         output_positional_encodings = []
 
@@ -977,8 +975,8 @@ class EdgeTamVideoPerceiverResampler(nn.Module):
     def _forward_1d(
         self,
         hidden_states: torch.Tensor,
-        positional_encoding: Optional[torch.Tensor] = None,
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        positional_encoding: torch.Tensor | None = None,
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
         batch_size = hidden_states.shape[0]
 
         latents = self.latents_1d.unsqueeze(0).expand(batch_size, -1, -1)
@@ -1247,8 +1245,8 @@ class EdgeTamVideoModel(Sam2VideoModel):
     def forward(
         self,
         inference_session: EdgeTamVideoInferenceSession,
-        frame_idx: Optional[int] = None,
-        frame: Optional[torch.Tensor] = None,
+        frame_idx: int | None = None,
+        frame: torch.Tensor | None = None,
         reverse: bool = False,
         **kwargs,
     ) -> EdgeTamVideoSegmentationOutput:
@@ -1394,11 +1392,11 @@ class EdgeTamVideoModel(Sam2VideoModel):
         obj_idx: int,
         batch_size: int,
         is_init_cond_frame: bool,
-        point_inputs: Optional[torch.Tensor],
-        mask_inputs: Optional[torch.Tensor],
+        point_inputs: torch.Tensor | None,
+        mask_inputs: torch.Tensor | None,
         reverse: bool,
         run_mem_encoder: bool,
-        prev_sam_mask_logits: Optional[torch.Tensor] = None,
+        prev_sam_mask_logits: torch.Tensor | None = None,
         streaming: bool = False,
     ) -> dict[str, Any]:
         """

transformers/models/efficientloftr/configuration_efficientloftr.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Optional
 
 from ...configuration_utils import PreTrainedConfig
 
@@ -97,9 +96,9 @@ class EfficientLoFTRConfig(PreTrainedConfig):
 
     def __init__(
         self,
-        stage_num_blocks: Optional[list[int]] = None,
-        out_features: Optional[list[int]] = None,
-        stage_stride: Optional[list[int]] = None,
+        stage_num_blocks: list[int] | None = None,
+        out_features: list[int] | None = None,
+        stage_stride: list[int] | None = None,
         hidden_size: int = 256,
         activation_function: str = "relu",
         q_aggregation_kernel_size: int = 4,
@@ -117,7 +116,7 @@ class EfficientLoFTRConfig(PreTrainedConfig):
         coarse_matching_border_removal: int = 2,
         fine_kernel_size: int = 8,
         batch_norm_eps: float = 1e-5,
-        rope_parameters: Optional[dict] = None,
+        rope_parameters: dict | None = None,
         fine_matching_slice_dim: int = 8,
         fine_matching_regress_temperature: float = 10.0,
         initializer_range: float = 0.02,

transformers/models/efficientloftr/image_processing_efficientloftr.py

@@ -13,8 +13,6 @@
 # limitations under the License.
 """Image processor class for SuperPoint."""
 
-from typing import Optional, Union
-
 import numpy as np
 
 from ... import is_torch_available, is_vision_available
@@ -62,7 +60,7 @@ class EfficientLoFTRImageProcessorKwargs(ImagesKwargs, total=False):
 # Copied from transformers.models.superpoint.image_processing_superpoint.is_grayscale
 def is_grayscale(
     image: np.ndarray,
-    input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    input_data_format: str | ChannelDimension | None = None,
 ):
     if input_data_format == ChannelDimension.FIRST:
         if image.shape[0] == 1:
@@ -77,7 +75,7 @@ def is_grayscale(
 # Copied from transformers.models.superpoint.image_processing_superpoint.convert_to_grayscale
 def convert_to_grayscale(
     image: ImageInput,
-    input_data_format: Optional[Union[str, ChannelDimension]] = None,
+    input_data_format: str | ChannelDimension | None = None,
 ) -> ImageInput:
     """
     Converts an image to grayscale format using the NTSC formula. Only support numpy and PIL Image.
@@ -170,7 +168,7 @@ class EfficientLoFTRImageProcessor(BaseImageProcessor):
     def __init__(
         self,
         do_resize: bool = True,
-        size: Optional[dict[str, int]] = None,
+        size: dict[str, int] | None = None,
         resample: PILImageResampling = PILImageResampling.BILINEAR,
         do_rescale: bool = True,
         rescale_factor: float = 1 / 255,
@@ -193,8 +191,8 @@ class EfficientLoFTRImageProcessor(BaseImageProcessor):
         self,
         image: np.ndarray,
         size: dict[str, int],
-        data_format: Optional[Union[str, ChannelDimension]] = None,
-        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        data_format: str | ChannelDimension | None = None,
+        input_data_format: str | ChannelDimension | None = None,
         **kwargs,
     ):
         """
@@ -232,15 +230,15 @@ class EfficientLoFTRImageProcessor(BaseImageProcessor):
     def preprocess(
         self,
         images,
-        do_resize: Optional[bool] = None,
-        size: Optional[dict[str, int]] = None,
-        resample: Optional[PILImageResampling] = None,
-        do_rescale: Optional[bool] = None,
-        rescale_factor: Optional[float] = None,
-        do_grayscale: Optional[bool] = None,
-        return_tensors: Optional[Union[str, TensorType]] = None,
+        do_resize: bool | None = None,
+        size: dict[str, int] | None = None,
+        resample: PILImageResampling | None = None,
+        do_rescale: bool | None = None,
+        rescale_factor: float | None = None,
+        do_grayscale: bool | None = None,
+        return_tensors: str | TensorType | None = None,
         data_format: ChannelDimension = ChannelDimension.FIRST,
-        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: str | ChannelDimension | None = None,
         **kwargs,
     ) -> BatchFeature:
         """
@@ -345,7 +343,7 @@ class EfficientLoFTRImageProcessor(BaseImageProcessor):
     def post_process_keypoint_matching(
         self,
         outputs: "EfficientLoFTRKeypointMatchingOutput",
-        target_sizes: Union[TensorType, list[tuple]],
+        target_sizes: TensorType | list[tuple],
         threshold: float = 0.0,
     ) -> list[dict[str, torch.Tensor]]:
         """

transformers/models/efficientloftr/image_processing_efficientloftr_fast.py

@@ -4,11 +4,11 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_efficientloftr.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-from typing import Optional, Union
+from typing import Optional
 
 import torch
+import torchvision.transforms.v2.functional as tvF
 from PIL import Image, ImageDraw
-from torchvision.transforms.v2 import functional as F
 
 from ...image_processing_utils_fast import BaseImageProcessorFast, BatchFeature
 from ...image_transforms import group_images_by_shape, reorder_images
@@ -84,7 +84,7 @@ def convert_to_grayscale(
     """
     if is_grayscale(image):
         return image
-    return F.rgb_to_grayscale(image, num_output_channels=3)
+    return tvF.rgb_to_grayscale(image, num_output_channels=3)
 
 
 @auto_docstring
@@ -111,19 +111,20 @@ class EfficientLoFTRImageProcessorFast(BaseImageProcessorFast):
         **kwargs,
     ) -> ImageInput:
         # we need to handle image pairs validation and flattening
+        images = self.fetch_images(images)
         return flatten_pair_images(images)
 
     def _preprocess(
         self,
         images: list["torch.Tensor"],
-        size: Union[dict[str, int], SizeDict],
+        size: dict[str, int] | SizeDict,
         rescale_factor: float,
         do_rescale: bool,
         do_resize: bool,
-        interpolation: Optional["F.InterpolationMode"],
+        interpolation: Optional["tvF.InterpolationMode"],
         do_grayscale: bool,
         disable_grouping: bool,
-        return_tensors: Union[str, TensorType],
+        return_tensors: str | TensorType,
         **kwargs,
     ) -> BatchFeature:
         grouped_images, grouped_images_index = group_images_by_shape(images, disable_grouping=disable_grouping)
@@ -159,7 +160,7 @@ class EfficientLoFTRImageProcessorFast(BaseImageProcessorFast):
     def post_process_keypoint_matching(
         self,
         outputs: "EfficientLoFTRKeypointMatchingOutput",
-        target_sizes: Union[TensorType, list[tuple]],
+        target_sizes: TensorType | list[tuple],
         threshold: float = 0.0,
     ) -> list[dict[str, torch.Tensor]]:
         """

transformers/models/efficientloftr/modeling_efficientloftr.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 from collections.abc import Callable
 from dataclasses import dataclass
-from typing import Optional, Union
+from typing import Optional
 
 import torch
 from torch import nn
@@ -64,12 +64,12 @@ class EfficientLoFTRKeypointMatchingOutput(ModelOutput):
         num_keypoints)`, returned when `output_attentions=True` is passed or when `config.output_attentions=True`)
     """
 
-    loss: Optional[torch.FloatTensor] = None
-    matches: Optional[torch.FloatTensor] = None
-    matching_scores: Optional[torch.FloatTensor] = None
-    keypoints: Optional[torch.FloatTensor] = None
-    hidden_states: Optional[tuple[torch.FloatTensor]] = None
-    attentions: Optional[tuple[torch.FloatTensor]] = None
+    loss: torch.FloatTensor | None = None
+    matches: torch.FloatTensor | None = None
+    matching_scores: torch.FloatTensor | None = None
+    keypoints: torch.FloatTensor | None = None
+    hidden_states: tuple[torch.FloatTensor] | None = None
+    attentions: tuple[torch.FloatTensor] | None = None
 
 
 @compile_compatible_method_lru_cache(maxsize=32)
@@ -108,9 +108,9 @@ class EfficientLoFTRRotaryEmbedding(nn.Module):
     @staticmethod
     # Ignore copy
     def compute_default_rope_parameters(
-        config: Optional[EfficientLoFTRConfig] = None,
+        config: EfficientLoFTRConfig | None = None,
         device: Optional["torch.device"] = None,
-        seq_len: Optional[int] = None,
+        seq_len: int | None = None,
     ) -> tuple["torch.Tensor", float]:
         """
         Computes the inverse frequencies according to the original RoPE implementation
@@ -141,7 +141,7 @@ class EfficientLoFTRRotaryEmbedding(nn.Module):
     # Ignore copy
     @torch.no_grad()
     def forward(
-        self, x: torch.Tensor, position_ids: Optional[torch.LongTensor] = None, layer_type=None
+        self, x: torch.Tensor, position_ids: torch.LongTensor | None = None, layer_type=None
     ) -> tuple[torch.Tensor, torch.Tensor]:
         feats_height, feats_width = x.shape[-2:]
         embed_height = (feats_height - self.config.q_aggregation_kernel_size) // self.config.q_aggregation_stride + 1
@@ -276,7 +276,7 @@ class EfficientLoFTRAggregationLayer(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_hidden_states: torch.Tensor | None = None,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         query_states = hidden_states
         is_cross_attention = encoder_hidden_states is not None
@@ -301,7 +301,7 @@ def rotate_half(x):
 
 
 # Copied from transformers.models.cohere.modeling_cohere.apply_rotary_pos_emb
-def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
     """Applies Rotary Position Embedding to the query and key tensors.
 
     Args:
@@ -309,8 +309,6 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
         k (`torch.Tensor`): The key tensor.
         cos (`torch.Tensor`): The cosine part of the rotary embedding.
         sin (`torch.Tensor`): The sine part of the rotary embedding.
-        position_ids (`torch.Tensor`, *optional*):
-            Deprecated and unused.
         unsqueeze_dim (`int`, *optional*, defaults to 1):
             The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
             sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
@@ -350,7 +348,7 @@ def eager_attention_forward(
     query: torch.Tensor,
     key: torch.Tensor,
     value: torch.Tensor,
-    attention_mask: Optional[torch.Tensor],
+    attention_mask: torch.Tensor | None,
     scaling: float,
     dropout: float = 0.0,
     **kwargs: Unpack[TransformersKwargs],
@@ -400,10 +398,10 @@ class EfficientLoFTRAttention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        encoder_hidden_states: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+    ) -> tuple[torch.Tensor, torch.Tensor | None]:
         batch_size, seq_len, dim = hidden_states.shape
         input_shape = hidden_states.shape[:-1]
 
@@ -420,9 +418,9 @@ class EfficientLoFTRAttention(nn.Module):
         query_states = query_states.view(batch_size, seq_len, -1, self.head_dim).transpose(1, 2)
         key_states = key_states.view(batch_size, seq_len, -1, self.head_dim).transpose(1, 2)
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -470,8 +468,8 @@ class EfficientLoFTRAggregatedAttention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        encoder_hidden_states: torch.Tensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> torch.Tensor:
         batch_size, embed_dim, _, _ = hidden_states.shape
@@ -519,7 +517,7 @@ class EfficientLoFTRLocalFeatureTransformerLayer(GradientCheckpointingLayer):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> torch.Tensor:
         batch_size, _, embed_dim, height, width = hidden_states.shape
@@ -554,7 +552,7 @@ class EfficientLoFTRLocalFeatureTransformer(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> torch.Tensor:
         for layer in self.layers:
@@ -740,7 +738,7 @@ class EfficientLoFTRModel(EfficientLoFTRPreTrainedModel):
     def forward(
         self,
         pixel_values: torch.FloatTensor,
-        labels: Optional[torch.LongTensor] = None,
+        labels: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> BackboneOutput:
         r"""
@@ -750,12 +748,17 @@ class EfficientLoFTRModel(EfficientLoFTRPreTrainedModel):
         >>> from transformers import AutoImageProcessor, AutoModel
         >>> import torch
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
 
         >>> url = "https://github.com/magicleap/SuperGluePretrainedNetwork/blob/master/assets/phototourism_sample_images/london_bridge_78916675_4568141288.jpg?raw=true"
-        >>> image1 = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image1 = Image.open(BytesIO(response.read()))
+
         >>> url = "https://github.com/magicleap/SuperGluePretrainedNetwork/blob/master/assets/phototourism_sample_images/london_bridge_19481797_2295892421.jpg?raw=true"
-        >>> image2 = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image2 = Image.open(BytesIO(response.read()))
+
         >>> images = [image1, image2]
 
         >>> processor = AutoImageProcessor.from_pretrained("zju-community/efficient_loftr")
@@ -798,7 +801,7 @@ class EfficientLoFTRModel(EfficientLoFTRPreTrainedModel):
         return BackboneOutput(feature_maps=features)
 
 
-def mask_border(tensor: torch.Tensor, border_margin: int, value: Union[bool, float, int]) -> torch.Tensor:
+def mask_border(tensor: torch.Tensor, border_margin: int, value: bool | float | int) -> torch.Tensor:
     """
     Mask a tensor border with a given value
 
@@ -830,11 +833,11 @@ def mask_border(tensor: torch.Tensor, border_margin: int, value: Union[bool, flo
 
 
 def create_meshgrid(
-    height: Union[int, torch.Tensor],
-    width: Union[int, torch.Tensor],
+    height: int | torch.Tensor,
+    width: int | torch.Tensor,
     normalized_coordinates: bool = False,
-    device: Optional[torch.device] = None,
-    dtype: Optional[torch.dtype] = None,
+    device: torch.device | None = None,
+    dtype: torch.dtype | None = None,
 ) -> torch.Tensor:
     """
     Copied from kornia library : kornia/kornia/utils/grid.py:26
@@ -1307,7 +1310,7 @@ class EfficientLoFTRForKeypointMatching(EfficientLoFTRPreTrainedModel):
     def forward(
         self,
         pixel_values: torch.FloatTensor,
-        labels: Optional[torch.LongTensor] = None,
+        labels: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> EfficientLoFTRKeypointMatchingOutput:
         r"""
@@ -1317,12 +1320,17 @@ class EfficientLoFTRForKeypointMatching(EfficientLoFTRPreTrainedModel):
         >>> from transformers import AutoImageProcessor, AutoModel
         >>> import torch
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
 
         >>> url = "https://github.com/magicleap/SuperGluePretrainedNetwork/blob/master/assets/phototourism_sample_images/london_bridge_78916675_4568141288.jpg?raw=true"
-        >>> image1 = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image1 = Image.open(BytesIO(response.read()))
+
         >>> url = "https://github.com/magicleap/SuperGluePretrainedNetwork/blob/master/assets/phototourism_sample_images/london_bridge_19481797_2295892421.jpg?raw=true"
-        >>> image2 = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image2 = Image.open(BytesIO(response.read()))
+
         >>> images = [image1, image2]
 
         >>> processor = AutoImageProcessor.from_pretrained("zju-community/efficient_loftr")

transformers/models/efficientloftr/modular_efficientloftr.py

@@ -1,5 +1,3 @@
-from typing import Union
-
 import torch
 
 from ...utils import TensorType
@@ -11,7 +9,7 @@ class EfficientLoFTRImageProcessorFast(SuperGlueImageProcessorFast):
     def post_process_keypoint_matching(
         self,
         outputs: "EfficientLoFTRKeypointMatchingOutput",
-        target_sizes: Union[TensorType, list[tuple]],
+        target_sizes: TensorType | list[tuple],
         threshold: float = 0.0,
     ) -> list[dict[str, torch.Tensor]]:
         """

transformers/models/efficientnet/configuration_efficientnet.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2023 Google Research, Inc. and The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");

transformers/models/efficientnet/image_processing_efficientnet.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2023 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,8 +13,6 @@
 # limitations under the License.
 """Image processor class for EfficientNet."""
 
-from typing import Optional, Union
-
 import numpy as np
 
 from ...image_processing_utils import BaseImageProcessor, BatchFeature, get_size_dict
@@ -101,16 +98,16 @@ class EfficientNetImageProcessor(BaseImageProcessor):
     def __init__(
         self,
         do_resize: bool = True,
-        size: Optional[dict[str, int]] = None,
+        size: dict[str, int] | None = None,
         resample: PILImageResampling = PILImageResampling.BICUBIC,
         do_center_crop: bool = False,
-        crop_size: Optional[dict[str, int]] = None,
-        rescale_factor: Union[int, float] = 1 / 255,
+        crop_size: dict[str, int] | None = None,
+        rescale_factor: int | float = 1 / 255,
         rescale_offset: bool = False,
         do_rescale: bool = True,
         do_normalize: bool = True,
-        image_mean: Optional[Union[float, list[float]]] = None,
-        image_std: Optional[Union[float, list[float]]] = None,
+        image_mean: float | list[float] | None = None,
+        image_std: float | list[float] | None = None,
         include_top: bool = True,
         **kwargs,
     ) -> None:
@@ -138,8 +135,8 @@ class EfficientNetImageProcessor(BaseImageProcessor):
         image: np.ndarray,
         size: dict[str, int],
         resample: PILImageResampling = PILImageResampling.BICUBIC,
-        data_format: Optional[Union[str, ChannelDimension]] = None,
-        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        data_format: str | ChannelDimension | None = None,
+        input_data_format: str | ChannelDimension | None = None,
         **kwargs,
     ) -> np.ndarray:
         """
@@ -184,10 +181,10 @@ class EfficientNetImageProcessor(BaseImageProcessor):
     def rescale(
         self,
         image: np.ndarray,
-        scale: Union[int, float],
+        scale: int | float,
         offset: bool = True,
-        data_format: Optional[Union[str, ChannelDimension]] = None,
-        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        data_format: str | ChannelDimension | None = None,
+        input_data_format: str | ChannelDimension | None = None,
         **kwargs,
     ):
         """
@@ -225,21 +222,21 @@ class EfficientNetImageProcessor(BaseImageProcessor):
     def preprocess(
         self,
         images: ImageInput,
-        do_resize: Optional[bool] = None,
-        size: Optional[dict[str, int]] = None,
+        do_resize: bool | None = None,
+        size: dict[str, int] | None = None,
         resample=None,
-        do_center_crop: Optional[bool] = None,
-        crop_size: Optional[dict[str, int]] = None,
-        do_rescale: Optional[bool] = None,
-        rescale_factor: Optional[float] = None,
-        rescale_offset: Optional[bool] = None,
-        do_normalize: Optional[bool] = None,
-        image_mean: Optional[Union[float, list[float]]] = None,
-        image_std: Optional[Union[float, list[float]]] = None,
-        include_top: Optional[bool] = None,
-        return_tensors: Optional[Union[str, TensorType]] = None,
+        do_center_crop: bool | None = None,
+        crop_size: dict[str, int] | None = None,
+        do_rescale: bool | None = None,
+        rescale_factor: float | None = None,
+        rescale_offset: bool | None = None,
+        do_normalize: bool | None = None,
+        image_mean: float | list[float] | None = None,
+        image_std: float | list[float] | None = None,
+        include_top: bool | None = None,
+        return_tensors: str | TensorType | None = None,
         data_format: ChannelDimension = ChannelDimension.FIRST,
-        input_data_format: Optional[Union[str, ChannelDimension]] = None,
+        input_data_format: str | ChannelDimension | None = None,
     ) -> PIL.Image.Image:
         """
         Preprocess an image or batch of images.