Remove nested import logic for torchvision (#40940)

* remove nested import logic for torchvision

* remove unnecessary protected imports

* remove unnecessarry protected import in modular (and modeling)

* fix wrongly remove protected imports

Author: yonigozlan

src/transformers/image_processing_utils_fast.py

@@ -61,14 +61,14 @@
 
 if is_torchvision_available():
     from .image_utils import pil_torch_interpolation_mapping
-
-    if is_torchvision_v2_available():
-        from torchvision.transforms.v2 import functional as F
-    else:
-        from torchvision.transforms import functional as F
 else:
     pil_torch_interpolation_mapping = None
 
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+elif is_torchvision_available():
+    from torchvision.transforms import functional as F
+
 logger = logging.get_logger(__name__)
 
 

src/transformers/models/aria/modeling_aria.py

@@ -21,6 +21,9 @@
 from dataclasses import dataclass
 from typing import Callable, Optional, Union
 
+import torch
+from torch import nn
+
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
@@ -35,16 +38,10 @@
 from ...utils import TransformersKwargs, auto_docstring, can_return_tuple
 from ...utils.deprecation import deprecate_kwarg
 from ...utils.generic import check_model_inputs
-from ...utils.import_utils import is_torch_available
 from ..auto import AutoModel
 from .configuration_aria import AriaConfig, AriaTextConfig
 
 
-if is_torch_available():
-    import torch
-    from torch import nn
-
-
 @use_kernel_forward_from_hub("RMSNorm")
 class AriaTextRMSNorm(nn.Module):
     def __init__(self, hidden_size, eps=1e-6):

src/transformers/models/aria/modular_aria.py

@@ -16,6 +16,8 @@
 from typing import Optional, Union
 
 import numpy as np
+import torch
+from torch import nn
 
 from ...activations import ACT2FN
 from ...cache_utils import Cache
@@ -39,7 +41,6 @@
 from ...processing_utils import MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack
 from ...tokenization_utils import PreTokenizedInput, TextInput
 from ...utils import TensorType, TransformersKwargs, auto_docstring, can_return_tuple, logging
-from ...utils.import_utils import is_torch_available
 from ..auto import CONFIG_MAPPING, AutoConfig, AutoTokenizer
 from ..llama.configuration_llama import LlamaConfig
 from ..llama.modeling_llama import (
@@ -62,10 +63,6 @@
 
 logger = logging.get_logger(__name__)
 
-if is_torch_available():
-    import torch
-    from torch import nn
-
 
 def sequential_experts_gemm(token_states, expert_weights, tokens_per_expert):
     """

src/transformers/models/beit/image_processing_beit_fast.py

@@ -16,6 +16,8 @@
 
 from typing import Optional, Union
 
+import torch
+
 from ...image_processing_utils import BatchFeature
 from ...image_processing_utils_fast import (
     BaseImageProcessorFast,
@@ -36,18 +38,13 @@
 from ...utils import (
     TensorType,
     auto_docstring,
-    is_torch_available,
-    is_torchvision_available,
     is_torchvision_v2_available,
 )
 
 
-if is_torch_available():
-    import torch
-
 if is_torchvision_v2_available():
     from torchvision.transforms.v2 import functional as F
-elif is_torchvision_available():
+else:
     from torchvision.transforms import functional as F
 
 

src/transformers/models/bridgetower/image_processing_bridgetower_fast.py

@@ -17,6 +17,8 @@
 from collections.abc import Iterable
 from typing import Optional, Union
 
+import torch
+
 from ...image_processing_utils_fast import (
     BaseImageProcessorFast,
     BatchFeature,
@@ -29,17 +31,13 @@
     reorder_images,
 )
 from ...image_utils import OPENAI_CLIP_MEAN, OPENAI_CLIP_STD, PILImageResampling
-from ...utils import auto_docstring, is_torch_available, is_torchvision_available, is_torchvision_v2_available
-
+from ...utils import auto_docstring, is_torchvision_v2_available
 
-if is_torch_available():
-    import torch
 
-if is_torchvision_available():
-    if is_torchvision_v2_available():
-        from torchvision.transforms.v2 import functional as F
-    else:
-        from torchvision.transforms import functional as F
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+else:
+    from torchvision.transforms import functional as F
 
 
 def make_pixel_mask(

src/transformers/models/chameleon/image_processing_chameleon_fast.py

@@ -17,28 +17,18 @@
 from typing import Optional
 
 import numpy as np
+import PIL
+import torch
 
 from ...image_processing_utils_fast import BaseImageProcessorFast
 from ...image_utils import ImageInput, PILImageResampling, SizeDict
-from ...utils import (
-    auto_docstring,
-    is_torch_available,
-    is_torchvision_available,
-    is_torchvision_v2_available,
-    is_vision_available,
-    logging,
-)
-
-
-if is_vision_available():
-    import PIL
-if is_torch_available():
-    import torch
-if is_torchvision_available():
-    if is_torchvision_v2_available():
-        from torchvision.transforms.v2 import functional as F
-    else:
-        from torchvision.transforms import functional as F
+from ...utils import auto_docstring, is_torchvision_v2_available, logging
+
+
+if is_torchvision_v2_available():
+    from torchvision.transforms.v2 import functional as F
+else:
+    from torchvision.transforms import functional as F
 
 logger = logging.get_logger(__name__)
 

src/transformers/models/cohere2_vision/modular_cohere2_vision.py

@@ -32,11 +32,7 @@
 from ...cache_utils import Cache
 from ...modeling_flash_attention_utils import FlashAttentionKwargs
 from ...processing_utils import Unpack
-from ...utils import (
-    TransformersKwargs,
-    auto_docstring,
-    logging,
-)
+from ...utils import TransformersKwargs, auto_docstring, logging
 from ...utils.generic import check_model_inputs
 from .configuration_cohere2_vision import Cohere2VisionConfig
 

src/transformers/models/colpali/modular_colpali.py

@@ -28,7 +28,6 @@
 if is_torch_available():
     import torch
 
-
 logger = logging.get_logger(__name__)
 
 

src/transformers/models/colqwen2/modular_colqwen2.py

@@ -30,7 +30,6 @@
 if is_torch_available():
     import torch
 
-
 logger = logging.get_logger(__name__)
 
 

src/transformers/models/conditional_detr/image_processing_conditional_detr_fast.py

@@ -7,6 +7,10 @@
 import pathlib
 from typing import Any, Optional, Union
 
+import torch
+from torch import nn
+from torchvision.io import read_image
+
 from ...image_processing_utils import BatchFeature, get_size_dict
 from ...image_processing_utils_fast import (
     BaseImageProcessorFast,
@@ -29,14 +33,7 @@
     validate_annotations,
 )
 from ...processing_utils import Unpack
-from ...utils import (
-    TensorType,
-    auto_docstring,
-    is_torch_available,
-    is_torchvision_available,
-    is_torchvision_v2_available,
-    logging,
-)
+from ...utils import TensorType, auto_docstring, is_torchvision_v2_available, logging
 from ...utils.import_utils import requires
 from .image_processing_conditional_detr import (
     compute_segments,
@@ -46,20 +43,9 @@
 )
 
 
-if is_torch_available():
-    import torch
-
-
-if is_torch_available():
-    from torch import nn
-
-
 if is_torchvision_v2_available():
-    from torchvision.io import read_image
     from torchvision.transforms.v2 import functional as F
-
-elif is_torchvision_available():
-    from torchvision.io import read_image
+else:
     from torchvision.transforms import functional as F