transformers 5.0.0rc2__py3-none-any.whl → 5.0.0rc3__py3-none-any.whl

transformers/models/sam_hq/modeling_sam_hq.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_sam_hq.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2025 Google Inc. HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -22,7 +21,6 @@
 import collections
 from collections.abc import Callable
 from dataclasses import dataclass
-from typing import Optional, Union
 
 import numpy as np
 import torch
@@ -62,12 +60,12 @@ class SamHQVisionEncoderOutput(ModelOutput):
         This is specific to SAM-HQ and not present in base SAM.
     """
 
-    image_embeds: Optional[torch.FloatTensor] = None
-    last_hidden_state: Optional[torch.FloatTensor] = None
-    hidden_states: Optional[tuple[torch.FloatTensor, ...]] = None
-    attentions: Optional[tuple[torch.FloatTensor, ...]] = None
+    image_embeds: torch.FloatTensor | None = None
+    last_hidden_state: torch.FloatTensor | None = None
+    hidden_states: tuple[torch.FloatTensor, ...] | None = None
+    attentions: tuple[torch.FloatTensor, ...] | None = None
 
-    intermediate_embeddings: Optional[list[torch.FloatTensor]] = None
+    intermediate_embeddings: list[torch.FloatTensor] | None = None
 
 
 @dataclass
@@ -83,8 +81,8 @@ class SamHQMMaskDecoderOutputs(ModelOutput):
     """
 
     masks: torch.FloatTensor
-    iou_scores: Optional[torch.FloatTensor] = None
-    mask_decoder_attentions: Optional[torch.FloatTensor] = None
+    iou_scores: torch.FloatTensor | None = None
+    mask_decoder_attentions: torch.FloatTensor | None = None
 
 
 @dataclass
@@ -118,11 +116,11 @@ class SamHQImageSegmentationOutput(ModelOutput):
         heads.
     """
 
-    iou_scores: Optional[torch.FloatTensor] = None
-    pred_masks: Optional[torch.FloatTensor] = None
-    vision_hidden_states: Optional[tuple[torch.FloatTensor, ...]] = None
-    vision_attentions: Optional[tuple[torch.FloatTensor, ...]] = None
-    mask_decoder_attentions: Optional[tuple[torch.FloatTensor, ...]] = None
+    iou_scores: torch.FloatTensor | None = None
+    pred_masks: torch.FloatTensor | None = None
+    vision_hidden_states: tuple[torch.FloatTensor, ...] | None = None
+    vision_attentions: tuple[torch.FloatTensor, ...] | None = None
+    mask_decoder_attentions: tuple[torch.FloatTensor, ...] | None = None
 
 
 class SamHQVisionAttention(nn.Module):
@@ -557,8 +555,8 @@ class SamHQVisionEncoder(SamHQPreTrainedModel):
 
     @check_model_inputs(tie_last_hidden_states=False)
     def forward(
-        self, pixel_values: Optional[torch.FloatTensor] = None, **kwargs: Unpack[TransformersKwargs]
-    ) -> Union[tuple, SamHQVisionEncoderOutput]:
+        self, pixel_values: torch.FloatTensor | None = None, **kwargs: Unpack[TransformersKwargs]
+    ) -> tuple | SamHQVisionEncoderOutput:
         if pixel_values is None:
             raise ValueError("You have to specify pixel_values")
 
@@ -614,7 +612,7 @@ 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,
@@ -672,7 +670,7 @@ class SamHQAttention(nn.Module):
         query: Tensor,
         key: Tensor,
         value: Tensor,
-        attention_similarity: Optional[Tensor] = None,
+        attention_similarity: Tensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> Tensor:
         # Input projections
@@ -809,7 +807,7 @@ class SamHQTwoWayTransformer(nn.Module):
         attention_similarity: Tensor,
         target_embedding=None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, BaseModelOutput]:
+    ) -> tuple | BaseModelOutput:
         if image_embeddings is None:
             raise ValueError("You have to specify an image_embedding")
 
@@ -922,9 +920,9 @@ class SamHQMaskDecoder(nn.Module):
         dense_prompt_embeddings: torch.Tensor,
         multimask_output: bool,
         hq_token_only: bool,
-        intermediate_embeddings: Optional[list[torch.Tensor]] = None,
-        attention_similarity: Optional[torch.Tensor] = None,
-        target_embedding: Optional[torch.Tensor] = None,
+        intermediate_embeddings: list[torch.Tensor] | None = None,
+        attention_similarity: torch.Tensor | None = None,
+        target_embedding: torch.Tensor | None = None,
     ) -> SamHQMMaskDecoderOutputs:
         """
         Predict high-quality masks given image and prompt embeddings.
@@ -1089,9 +1087,9 @@ class SamHQVisionModel(SamHQPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        pixel_values: Optional[torch.FloatTensor] = None,
+        pixel_values: torch.FloatTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, SamHQVisionEncoderOutput]:
+    ) -> tuple | SamHQVisionEncoderOutput:
         return self.vision_encoder(pixel_values, **kwargs)
 
 
@@ -1186,10 +1184,10 @@ class SamHQPromptEncoder(nn.Module):
 
     def forward(
         self,
-        input_points: Optional[tuple[torch.Tensor, torch.Tensor]],
-        input_labels: Optional[torch.Tensor],
-        input_boxes: Optional[torch.Tensor],
-        input_masks: Optional[torch.Tensor],
+        input_points: tuple[torch.Tensor, torch.Tensor] | None,
+        input_labels: torch.Tensor | None,
+        input_boxes: torch.Tensor | None,
+        input_masks: torch.Tensor | None,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Embeds different types of prompts, returning both sparse and dense embeddings.
@@ -1284,10 +1282,10 @@ class SamHQModel(SamHQPreTrainedModel):
     @torch.no_grad()
     def get_prompt_embeddings(
         self,
-        input_points: Optional[torch.FloatTensor] = None,
-        input_labels: Optional[torch.LongTensor] = None,
-        input_boxes: Optional[torch.FloatTensor] = None,
-        input_masks: Optional[torch.LongTensor] = None,
+        input_points: torch.FloatTensor | None = None,
+        input_labels: torch.LongTensor | None = None,
+        input_boxes: torch.FloatTensor | None = None,
+        input_masks: torch.LongTensor | None = None,
     ):
         r"""
         Returns the prompt embeddings by passing the input points, labels, boxes and masks through the prompt encoder.
@@ -1318,17 +1316,17 @@ class SamHQModel(SamHQPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        pixel_values: Optional[torch.FloatTensor] = None,
-        input_points: Optional[torch.FloatTensor] = None,
-        input_labels: Optional[torch.LongTensor] = None,
-        input_boxes: Optional[torch.FloatTensor] = None,
-        input_masks: Optional[torch.LongTensor] = None,
-        image_embeddings: Optional[torch.FloatTensor] = None,
+        pixel_values: torch.FloatTensor | None = None,
+        input_points: torch.FloatTensor | None = None,
+        input_labels: torch.LongTensor | None = None,
+        input_boxes: torch.FloatTensor | None = None,
+        input_masks: torch.LongTensor | None = None,
+        image_embeddings: torch.FloatTensor | None = None,
         multimask_output: bool = True,
         hq_token_only: bool = False,
-        attention_similarity: Optional[torch.FloatTensor] = None,
-        target_embedding: Optional[torch.FloatTensor] = None,
-        intermediate_embeddings: Optional[list[torch.FloatTensor]] = None,
+        attention_similarity: torch.FloatTensor | None = None,
+        target_embedding: torch.FloatTensor | None = None,
+        intermediate_embeddings: list[torch.FloatTensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> list[dict[str, torch.Tensor]]:
         r"""

transformers/models/sam_hq/modular_sam_hq.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 Google Inc. HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -14,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from dataclasses import dataclass
-from typing import Optional, Union
 
 import torch
 from torch import nn
@@ -149,7 +147,7 @@ class SamHQVisionEncoderOutput(SamVisionEncoderOutput):
         This is specific to SAM-HQ and not present in base SAM.
     """
 
-    intermediate_embeddings: Optional[list[torch.FloatTensor]] = None
+    intermediate_embeddings: list[torch.FloatTensor] | None = None
 
 
 @dataclass
@@ -165,8 +163,8 @@ class SamHQMMaskDecoderOutputs(ModelOutput):
     """
 
     masks: torch.FloatTensor
-    iou_scores: Optional[torch.FloatTensor] = None
-    mask_decoder_attentions: Optional[torch.FloatTensor] = None
+    iou_scores: torch.FloatTensor | None = None
+    mask_decoder_attentions: torch.FloatTensor | None = None
 
 
 class SamHQImageSegmentationOutput(SamImageSegmentationOutput):
@@ -193,8 +191,8 @@ class SamHQVisionEncoder(SamVisionEncoder, SamHQPreTrainedModel):
 
     @check_model_inputs(tie_last_hidden_states=False)
     def forward(
-        self, pixel_values: Optional[torch.FloatTensor] = None, **kwargs: Unpack[TransformersKwargs]
-    ) -> Union[tuple, SamHQVisionEncoderOutput]:
+        self, pixel_values: torch.FloatTensor | None = None, **kwargs: Unpack[TransformersKwargs]
+    ) -> tuple | SamHQVisionEncoderOutput:
         if pixel_values is None:
             raise ValueError("You have to specify pixel_values")
 
@@ -285,9 +283,9 @@ class SamHQMaskDecoder(nn.Module):
         dense_prompt_embeddings: torch.Tensor,
         multimask_output: bool,
         hq_token_only: bool,
-        intermediate_embeddings: Optional[list[torch.Tensor]] = None,
-        attention_similarity: Optional[torch.Tensor] = None,
-        target_embedding: Optional[torch.Tensor] = None,
+        intermediate_embeddings: list[torch.Tensor] | None = None,
+        attention_similarity: torch.Tensor | None = None,
+        target_embedding: torch.Tensor | None = None,
     ) -> SamHQMMaskDecoderOutputs:
         """
         Predict high-quality masks given image and prompt embeddings.
@@ -471,17 +469,17 @@ class SamHQModel(SamModel):
 
     def forward(
         self,
-        pixel_values: Optional[torch.FloatTensor] = None,
-        input_points: Optional[torch.FloatTensor] = None,
-        input_labels: Optional[torch.LongTensor] = None,
-        input_boxes: Optional[torch.FloatTensor] = None,
-        input_masks: Optional[torch.LongTensor] = None,
-        image_embeddings: Optional[torch.FloatTensor] = None,
+        pixel_values: torch.FloatTensor | None = None,
+        input_points: torch.FloatTensor | None = None,
+        input_labels: torch.LongTensor | None = None,
+        input_boxes: torch.FloatTensor | None = None,
+        input_masks: torch.LongTensor | None = None,
+        image_embeddings: torch.FloatTensor | None = None,
         multimask_output: bool = True,
         hq_token_only: bool = False,
-        attention_similarity: Optional[torch.FloatTensor] = None,
-        target_embedding: Optional[torch.FloatTensor] = None,
-        intermediate_embeddings: Optional[list[torch.FloatTensor]] = None,
+        attention_similarity: torch.FloatTensor | None = None,
+        target_embedding: torch.FloatTensor | None = None,
+        intermediate_embeddings: list[torch.FloatTensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> list[dict[str, torch.Tensor]]:
         r"""

transformers/models/sam_hq/{processing_samhq.py → processing_sam_hq.py}

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,28 +16,55 @@ Processor class for SAMHQ.
 """
 
 from copy import deepcopy
-from typing import Optional, Union
 
 import numpy as np
 
 from ...image_utils import ImageInput
 from ...processing_utils import ImagesKwargs, ProcessingKwargs, ProcessorMixin, Unpack
-from ...tokenization_utils_base import BatchEncoding, PreTokenizedInput, TextInput
-from ...utils import is_torch_available
+from ...tokenization_utils_base import BatchEncoding
+from ...utils import auto_docstring, is_torch_available
 
 
 if is_torch_available():
     import torch
 
-NestedList = list[Union[Optional[float | int], "NestedList"]]
+NestedList = list[float | int | None | list[float | int | None]]
 
 
 class SamHQImagesKwargs(ImagesKwargs, total=False):
-    segmentation_maps: Optional[ImageInput]
-    input_points: Optional[NestedList]
-    input_labels: Optional[NestedList]
-    input_boxes: Optional[NestedList]
-    point_pad_value: Optional[int]
+    """
+    segmentation_maps (`ImageInput`, *optional*):
+        Ground truth segmentation maps to process alongside the input images. These maps are used for training
+        or evaluation purposes and are resized and normalized to match the processed image dimensions.
+    input_points (`NestedList`, *optional*):
+        Input points for prompt-based segmentation. Should be a nested list with structure
+        `[image_level, object_level, point_level, [x, y]]` where each point is specified as `[x, y]` coordinates
+        in the original image space. Points are normalized to the target image size before being passed to the model.
+    input_labels (`NestedList`, *optional*):
+        Labels for the input points, indicating whether each point is a foreground (1) or background (0) point.
+        Should be a nested list with structure `[image_level, object_level, point_level]`. Must have the same
+        structure as `input_points` (excluding the coordinate dimension).
+    input_boxes (`NestedList`, *optional*):
+        Bounding boxes for prompt-based segmentation. Should be a nested list with structure
+        `[image_level, box_level, [x1, y1, x2, y2]]` where each box is specified as `[x1, y1, x2, y2]` coordinates
+        in the original image space. Boxes are normalized to the target image size before being passed to the model.
+    point_pad_value (`int`, *optional*, defaults to `None`):
+        The value used for padding input points when batching sequences of different lengths. This value marks
+        padded positions and is preserved during coordinate normalization to distinguish real points from padding.
+        If `None`, the default pad value from the processor configuration is used.
+    mask_size (`dict[str, int]`, *optional*):
+        Dictionary specifying the target mask size with keys `"height"` and `"width"`. This determines the
+        resolution of the output segmentation masks generated by the model.
+    mask_pad_size (`dict[str, int]`, *optional*):
+        Dictionary specifying the padding size for masks with keys `"height"` and `"width"`. This is used when
+        batching masks of different sizes to ensure consistent dimensions.
+    """
+
+    segmentation_maps: ImageInput | None
+    input_points: NestedList | None
+    input_labels: NestedList | None
+    input_boxes: NestedList | None
+    point_pad_value: int | None
     mask_size: dict[str, int]
     mask_pad_size: dict[str, int]
 
@@ -52,19 +78,8 @@ class SamHQProcessorKwargs(ProcessingKwargs, total=False):
     }
 
 
+@auto_docstring
 class SamHQProcessor(ProcessorMixin):
-    r"""
-    Constructs a SAM HQ processor which wraps a SAM  image processor and an 2D points & Bounding boxes processor into a
-    single processor.
-
-    [`SamHQProcessor`] offers all the functionalities of [`SamImageProcessor`]. See the docstring of
-    [`~SamImageProcessor.__call__`] for more information.
-
-    Args:
-        image_processor (`SamImageProcessor`):
-            An instance of [`SamImageProcessor`]. The image processor is a required input.
-    """
-
     def __init__(self, image_processor):
         super().__init__(image_processor)
         # Ensure image_processor is properly initialized
@@ -74,16 +89,12 @@ class SamHQProcessor(ProcessorMixin):
             raise ValueError("image_processor.size is not set")
         self.target_size = self.image_processor.size["longest_edge"]
 
+    @auto_docstring
     def __call__(
         self,
-        images: Optional[ImageInput] = None,
-        text: Optional[Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]]] = None,
+        images: ImageInput | None = None,
         **kwargs: Unpack[SamHQProcessorKwargs],
     ) -> BatchEncoding:
-        """
-        This method uses [`SamImageProcessor.__call__`] method to prepare image(s) for the model. It also prepares 2D
-        points and bounding boxes for the model if they are provided.
-        """
         output_kwargs = self._merge_kwargs(
             SamHQProcessorKwargs,
             tokenizer_init_kwargs={},

transformers/models/seamless_m4t/configuration_seamless_m4t.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");

transformers/models/seamless_m4t/feature_extraction_seamless_m4t.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2023 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -16,8 +15,6 @@
 Feature extractor class for SeamlessM4T
 """
 
-from typing import Optional, Union
-
 import numpy as np
 
 from ...utils import is_torch_available
@@ -143,15 +140,15 @@ class SeamlessM4TFeatureExtractor(SequenceFeatureExtractor):
 
     def __call__(
         self,
-        raw_speech: Union[np.ndarray, list[float], list[np.ndarray], list[list[float]]],
-        padding: Union[bool, str, PaddingStrategy] = True,
-        pad_to_multiple_of: Optional[int] = 2,
-        max_length: Optional[int] = None,
+        raw_speech: np.ndarray | list[float] | list[np.ndarray] | list[list[float]],
+        padding: bool | str | PaddingStrategy = True,
+        pad_to_multiple_of: int | None = 2,
+        max_length: int | None = None,
         truncation: bool = False,
-        return_tensors: Optional[Union[str, TensorType]] = None,
-        sampling_rate: Optional[int] = None,
-        return_attention_mask: Optional[bool] = None,
-        do_normalize_per_mel_bins: Optional[bool] = True,
+        return_tensors: str | TensorType | None = None,
+        sampling_rate: int | None = None,
+        return_attention_mask: bool | None = None,
+        do_normalize_per_mel_bins: bool | None = True,
         **kwargs,
     ) -> BatchFeature:
         """