transformers 5.0.0rc3__py3-none-any.whl → 5.1.0__py3-none-any.whl

transformers/models/owlv2/image_processing_owlv2_fast.py

@@ -22,7 +22,7 @@ import warnings
 from typing import TYPE_CHECKING, Optional
 
 import torch
-from torchvision.transforms.v2 import functional as F
+import torchvision.transforms.v2.functional as tvF
 
 from ...image_processing_utils_fast import BaseImageProcessorFast, BatchFeature
 from ...image_transforms import center_to_corners_format, group_images_by_shape, reorder_images
@@ -192,7 +192,7 @@ class Owlv2ImageProcessorFast(BaseImageProcessorFast):
         pad_right = size - width
 
         padding = (0, 0, pad_right, pad_bottom)
-        padded_image = F.pad(images, padding, fill=constant_value)
+        padded_image = tvF.pad(images, padding, fill=constant_value)
         return padded_image
 
     def pad(
@@ -264,14 +264,14 @@ class Owlv2ImageProcessorFast(BaseImageProcessorFast):
             else:
                 kernel_sizes = 2 * torch.ceil(3 * anti_aliasing_sigma).int() + 1
 
-                filtered = F.gaussian_blur(
+                filtered = tvF.gaussian_blur(
                     image, (kernel_sizes[0], kernel_sizes[1]), sigma=anti_aliasing_sigma.tolist()
                 )
 
         else:
             filtered = image
 
-        out = F.resize(filtered, size=(size.height, size.width), antialias=False)
+        out = tvF.resize(filtered, size=(size.height, size.width), antialias=False)
 
         return out
 
@@ -280,7 +280,7 @@ class Owlv2ImageProcessorFast(BaseImageProcessorFast):
         images: list["torch.Tensor"],
         do_resize: bool,
         size: SizeDict,
-        interpolation: Optional["F.InterpolationMode"],
+        interpolation: Optional["tvF.InterpolationMode"],
         do_pad: bool,
         do_rescale: bool,
         rescale_factor: float,

transformers/models/owlv2/modeling_owlv2.py

@@ -25,10 +25,12 @@ from ...modeling_attn_mask_utils import _create_4d_causal_attention_mask, _prepa
 from ...modeling_layers import GradientCheckpointingLayer
 from ...modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
 from ...modeling_utils import PreTrainedModel
+from ...processing_utils import Unpack
 from ...utils import (
     ModelOutput,
+    TransformersKwargs,
     auto_docstring,
-    filter_out_non_signature_kwargs,
+    can_return_tuple,
     is_vision_available,
     logging,
     torch_int,
@@ -912,13 +914,15 @@ class Owlv2VisionModel(Owlv2PreTrainedModel):
         Examples:
         ```python
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from transformers import AutoProcessor, Owlv2VisionModel
 
         >>> model = Owlv2VisionModel.from_pretrained("google/owlv2-base-patch16")
         >>> processor = AutoProcessor.from_pretrained("google/owlv2-base-patch16")
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
 
         >>> inputs = processor(images=image, return_tensors="pt")
 
@@ -972,23 +976,20 @@ class Owlv2Model(Owlv2PreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @filter_out_non_signature_kwargs()
+    @can_return_tuple
     @auto_docstring
     def get_text_features(
         self,
         input_ids: torch.Tensor,
         attention_mask: torch.Tensor | None = None,
-    ) -> torch.FloatTensor:
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple | BaseModelOutputWithPooling:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size * num_max_text_queries, sequence_length)`):
             Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See
             [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input
             IDs?](../glossary#input-ids)
 
-        Returns:
-            text_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The text embeddings obtained by
-            applying the projection layer to the pooled output of [`Owlv2TextModel`].
-
         Examples:
         ```python
         >>> import torch
@@ -1003,23 +1004,26 @@ class Owlv2Model(Owlv2PreTrainedModel):
         ...     text_features = model.get_text_features(**inputs)
         ```"""
         # Get embeddings for all text queries in all batch samples
-        text_outputs: BaseModelOutputWithPooling = self.text_model(input_ids=input_ids, attention_mask=attention_mask)
-        text_features = self.text_projection(text_outputs.pooler_output)
+        text_outputs: BaseModelOutputWithPooling = self.text_model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            return_dict=True,
+            **kwargs,
+        )
+        pooled_output = text_outputs.pooler_output
+        text_outputs.pooler_output = self.text_projection(pooled_output)
 
-        return text_features
+        return text_outputs
 
-    @filter_out_non_signature_kwargs()
+    @can_return_tuple
     @auto_docstring
     def get_image_features(
         self,
         pixel_values: torch.Tensor,
         interpolate_pos_encoding: bool = False,
-    ) -> torch.FloatTensor:
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple | BaseModelOutputWithPooling:
         r"""
-        Returns:
-            image_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The image embeddings obtained by
-            applying the projection layer to the pooled output of [`Owlv2VisionModel`].
-
         Examples:
         ```python
         >>> import torch
@@ -1039,10 +1043,12 @@ class Owlv2Model(Owlv2PreTrainedModel):
         vision_outputs: BaseModelOutputWithPooling = self.vision_model(
             pixel_values=pixel_values,
             interpolate_pos_encoding=interpolate_pos_encoding,
+            return_dict=True,
+            **kwargs,
         )
-        image_features = self.visual_projection(vision_outputs.pooler_output)
+        vision_outputs.pooler_output = self.visual_projection(vision_outputs.pooler_output)
 
-        return image_features
+        return vision_outputs
 
     @auto_docstring
     def forward(
@@ -1067,13 +1073,15 @@ class Owlv2Model(Owlv2PreTrainedModel):
         Examples:
         ```python
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from transformers import AutoProcessor, Owlv2Model
 
         >>> model = Owlv2Model.from_pretrained("google/owlv2-base-patch16-ensemble")
         >>> processor = AutoProcessor.from_pretrained("google/owlv2-base-patch16-ensemble")
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> inputs = processor(text=[["a photo of a cat", "a photo of a dog"]], images=image, return_tensors="pt")
         >>> outputs = model(**inputs)
         >>> logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
@@ -1492,7 +1500,8 @@ class Owlv2ForObjectDetection(Owlv2PreTrainedModel):
 
         Examples:
         ```python
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from PIL import Image
         >>> import torch
         >>> from transformers import AutoProcessor, Owlv2ForObjectDetection
@@ -1501,9 +1510,11 @@ class Owlv2ForObjectDetection(Owlv2PreTrainedModel):
         >>> model = Owlv2ForObjectDetection.from_pretrained("google/owlv2-base-patch16-ensemble")
 
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> query_url = "http://images.cocodataset.org/val2017/000000001675.jpg"
-        >>> query_image = Image.open(requests.get(query_url, stream=True).raw)
+        >>> with httpx.stream("GET", query_url) as response:
+        ...     query_image = Image.open(BytesIO(response.read()))
         >>> inputs = processor(images=image, query_images=query_image, return_tensors="pt")
 
         >>> # forward pass
@@ -1617,7 +1628,8 @@ class Owlv2ForObjectDetection(Owlv2PreTrainedModel):
 
         Examples:
         ```python
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from PIL import Image
         >>> import torch
 
@@ -1627,7 +1639,8 @@ class Owlv2ForObjectDetection(Owlv2PreTrainedModel):
         >>> model = Owlv2ForObjectDetection.from_pretrained("google/owlv2-base-patch16-ensemble")
 
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> text_labels = [["a photo of a cat", "a photo of a dog"]]
         >>> inputs = processor(text=text_labels, images=image, return_tensors="pt")
         >>> outputs = model(**inputs)

transformers/models/owlv2/modular_owlv2.py

@@ -17,7 +17,7 @@ import warnings
 from typing import Optional
 
 import torch
-from torchvision.transforms.v2 import functional as F
+import torchvision.transforms.v2.functional as tvF
 
 from ...image_processing_utils_fast import (
     BatchFeature,
@@ -61,7 +61,7 @@ class Owlv2ImageProcessorFast(OwlViTImageProcessorFast):
         pad_right = size - width
 
         padding = (0, 0, pad_right, pad_bottom)
-        padded_image = F.pad(images, padding, fill=constant_value)
+        padded_image = tvF.pad(images, padding, fill=constant_value)
         return padded_image
 
     def pad(
@@ -133,14 +133,14 @@ class Owlv2ImageProcessorFast(OwlViTImageProcessorFast):
             else:
                 kernel_sizes = 2 * torch.ceil(3 * anti_aliasing_sigma).int() + 1
 
-                filtered = F.gaussian_blur(
+                filtered = tvF.gaussian_blur(
                     image, (kernel_sizes[0], kernel_sizes[1]), sigma=anti_aliasing_sigma.tolist()
                 )
 
         else:
             filtered = image
 
-        out = F.resize(filtered, size=(size.height, size.width), antialias=False)
+        out = tvF.resize(filtered, size=(size.height, size.width), antialias=False)
 
         return out
 
@@ -149,7 +149,7 @@ class Owlv2ImageProcessorFast(OwlViTImageProcessorFast):
         images: list["torch.Tensor"],
         do_resize: bool,
         size: SizeDict,
-        interpolation: Optional["F.InterpolationMode"],
+        interpolation: Optional["tvF.InterpolationMode"],
         do_pad: bool,
         do_rescale: bool,
         rescale_factor: float,

transformers/models/owlvit/configuration_owlvit.py

@@ -101,7 +101,10 @@ class OwlViTTextConfig(PreTrainedConfig):
         eos_token_id=49407,
         **kwargs,
     ):
-        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)
+        super().__init__(**kwargs)
+        self.pad_token_id = pad_token_id
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
 
         self.vocab_size = vocab_size
         self.hidden_size = hidden_size

transformers/models/owlvit/modeling_owlvit.py

@@ -25,10 +25,12 @@ from ...modeling_attn_mask_utils import _create_4d_causal_attention_mask, _prepa
 from ...modeling_layers import GradientCheckpointingLayer
 from ...modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
 from ...modeling_utils import PreTrainedModel
+from ...processing_utils import Unpack
 from ...utils import (
     ModelOutput,
+    TransformersKwargs,
     auto_docstring,
-    filter_out_non_signature_kwargs,
+    can_return_tuple,
     is_vision_available,
     logging,
     torch_int,
@@ -894,13 +896,15 @@ class OwlViTVisionModel(OwlViTPreTrainedModel):
         Examples:
         ```python
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from transformers import AutoProcessor, OwlViTVisionModel
 
         >>> model = OwlViTVisionModel.from_pretrained("google/owlvit-base-patch32")
         >>> processor = AutoProcessor.from_pretrained("google/owlvit-base-patch32")
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
 
         >>> inputs = processor(images=image, return_tensors="pt")
 
@@ -953,23 +957,20 @@ class OwlViTModel(OwlViTPreTrainedModel):
         # Initialize weights and apply final processing
         self.post_init()
 
-    @filter_out_non_signature_kwargs()
+    @can_return_tuple
     @auto_docstring
     def get_text_features(
         self,
         input_ids: torch.Tensor,
         attention_mask: torch.Tensor | None = None,
-    ) -> torch.FloatTensor:
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple | BaseModelOutputWithPooling:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size * num_max_text_queries, sequence_length)`):
             Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See
             [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input
             IDs?](../glossary#input-ids)
 
-        Returns:
-            text_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The text embeddings obtained by
-            applying the projection layer to the pooled output of [`OwlViTTextModel`].
-
         Examples:
         ```python
         >>> import torch
@@ -984,23 +985,26 @@ class OwlViTModel(OwlViTPreTrainedModel):
         ...     text_features = model.get_text_features(**inputs)
         ```"""
         # Get embeddings for all text queries in all batch samples
-        text_outputs: BaseModelOutputWithPooling = self.text_model(input_ids=input_ids, attention_mask=attention_mask)
-        text_features = self.text_projection(text_outputs.pooler_output)
+        text_outputs: BaseModelOutputWithPooling = self.text_model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            return_dict=True,
+            **kwargs,
+        )
+        pooled_output = text_outputs.pooler_output
+        text_outputs.pooler_output = self.text_projection(pooled_output)
 
-        return text_features
+        return text_outputs
 
-    @filter_out_non_signature_kwargs()
+    @can_return_tuple
     @auto_docstring
     def get_image_features(
         self,
         pixel_values: torch.Tensor,
         interpolate_pos_encoding: bool = False,
-    ) -> torch.FloatTensor:
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple | BaseModelOutputWithPooling:
         r"""
-        Returns:
-            image_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The image embeddings obtained by
-            applying the projection layer to the pooled output of [`OwlViTVisionModel`].
-
         Examples:
         ```python
         >>> import torch
@@ -1020,10 +1024,12 @@ class OwlViTModel(OwlViTPreTrainedModel):
         vision_outputs: BaseModelOutputWithPooling = self.vision_model(
             pixel_values=pixel_values,
             interpolate_pos_encoding=interpolate_pos_encoding,
+            return_dict=True,
+            **kwargs,
         )
-        image_features = self.visual_projection(vision_outputs.pooler_output)
+        vision_outputs.pooler_output = self.visual_projection(vision_outputs.pooler_output)
 
-        return image_features
+        return vision_outputs
 
     @auto_docstring
     def forward(
@@ -1048,13 +1054,15 @@ class OwlViTModel(OwlViTPreTrainedModel):
         Examples:
         ```python
         >>> from PIL import Image
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from transformers import AutoProcessor, OwlViTModel
 
         >>> model = OwlViTModel.from_pretrained("google/owlvit-base-patch32")
         >>> processor = AutoProcessor.from_pretrained("google/owlvit-base-patch32")
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> inputs = processor(text=[["a photo of a cat", "a photo of a dog"]], images=image, return_tensors="pt")
         >>> outputs = model(**inputs)
         >>> logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
@@ -1448,7 +1456,8 @@ class OwlViTForObjectDetection(OwlViTPreTrainedModel):
 
         Examples:
         ```python
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from PIL import Image
         >>> import torch
         >>> from transformers import AutoProcessor, OwlViTForObjectDetection
@@ -1456,9 +1465,11 @@ class OwlViTForObjectDetection(OwlViTPreTrainedModel):
         >>> processor = AutoProcessor.from_pretrained("google/owlvit-base-patch16")
         >>> model = OwlViTForObjectDetection.from_pretrained("google/owlvit-base-patch16")
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> query_url = "http://images.cocodataset.org/val2017/000000001675.jpg"
-        >>> query_image = Image.open(requests.get(query_url, stream=True).raw)
+        >>> with httpx.stream("GET", query_url) as response:
+        ...     query_image = Image.open(BytesIO(response.read()))
         >>> inputs = processor(images=image, query_images=query_image, return_tensors="pt")
         >>> with torch.no_grad():
         ...     outputs = model.image_guided_detection(**inputs)
@@ -1558,7 +1569,8 @@ class OwlViTForObjectDetection(OwlViTPreTrainedModel):
 
         Examples:
         ```python
-        >>> import requests
+        >>> import httpx
+        >>> from io import BytesIO
         >>> from PIL import Image
         >>> import torch
 
@@ -1568,7 +1580,8 @@ class OwlViTForObjectDetection(OwlViTPreTrainedModel):
         >>> model = OwlViTForObjectDetection.from_pretrained("google/owlvit-base-patch32")
 
         >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
-        >>> image = Image.open(requests.get(url, stream=True).raw)
+        >>> with httpx.stream("GET", url) as response:
+        ...     image = Image.open(BytesIO(response.read()))
         >>> text_labels = [["a photo of a cat", "a photo of a dog"]]
         >>> inputs = processor(text=text_labels, images=image, return_tensors="pt")
         >>> outputs = model(**inputs)

transformers/models/paddleocr_vl/configuration_paddleocr_vl.py

@@ -246,13 +246,11 @@ class PaddleOCRTextConfig(PreTrainedConfig):
         self.head_dim = head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads
         self.rope_parameters = rope_parameters
 
-        super().__init__(
-            pad_token_id=pad_token_id,
-            bos_token_id=bos_token_id,
-            eos_token_id=eos_token_id,
-            tie_word_embeddings=tie_word_embeddings,
-            **kwargs,
-        )
+        self.tie_word_embeddings = tie_word_embeddings
+        self.pad_token_id = pad_token_id
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        super().__init__(**kwargs)
 
 
 class PaddleOCRVLConfig(PreTrainedConfig):
@@ -279,6 +277,8 @@ class PaddleOCRVLConfig(PreTrainedConfig):
             The token index to denote start of vision input.
         vision_end_token_id (`int`, *optional*, defaults to 101306):
             The token index to denote end of vision input.
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether the model's input and output word embeddings should be tied.
 
     ```python
     >>> from transformers import PaddleOCRVLForConditionalGeneration, PaddleOCRVLConfig
@@ -306,6 +306,7 @@ class PaddleOCRVLConfig(PreTrainedConfig):
         video_token_id=100296,
         vision_start_token_id=101305,
         vision_end_token_id=101306,
+        tie_word_embeddings=True,
         **kwargs,
     ):
         if isinstance(vision_config, dict):
@@ -327,9 +328,7 @@ class PaddleOCRVLConfig(PreTrainedConfig):
         self.video_token_id = video_token_id
         self.vision_start_token_id = vision_start_token_id
         self.vision_end_token_id = vision_end_token_id
-
-        # FIXME: arthur/cyril - tying has to be used from the text config
-        kwargs["tie_word_embeddings"] = self.text_config.tie_word_embeddings
+        self.tie_word_embeddings = tie_word_embeddings
         super().__init__(**kwargs)