🚨 Delete deprecations with end-cycle in v4.xx and v5.0

src/transformers/models/aimv2/modeling_aimv2.py

@@ -37,7 +37,6 @@
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from ...processing_utils import Unpack
 from ...utils import ModelOutput, TransformersKwargs, auto_docstring, can_return_tuple, filter_out_non_signature_kwargs
-from ...utils.deprecation import deprecate_kwarg
 from ...utils.generic import check_model_inputs
 from .configuration_aimv2 import Aimv2Config, Aimv2TextConfig, Aimv2VisionConfig
 
@@ -444,13 +443,11 @@ def __init__(self, config: Aimv2VisionConfig):
     def get_input_embeddings(self) -> nn.Module:
         return self.embeddings.patch_embed
 
-    @deprecate_kwarg("attention_mask", version="v4.58.0")
     @check_model_inputs(tie_last_hidden_states=False)
     @auto_docstring
     def forward(
         self,
         pixel_values,
-        attention_mask: Optional[torch.Tensor] = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> BaseModelOutputWithPooling:
         r"""

src/transformers/models/aimv2/modular_aimv2.py

@@ -32,7 +32,6 @@
     auto_docstring,
     can_return_tuple,
 )
-from ...utils.deprecation import deprecate_kwarg
 from ...utils.generic import check_model_inputs
 from ..clip.modeling_clip import CLIPModel, CLIPTextEmbeddings, _get_vector_norm
 from ..llama.modeling_llama import LlamaMLP, LlamaRMSNorm
@@ -487,13 +486,11 @@ def __init__(self, config: Aimv2VisionConfig):
     def get_input_embeddings(self) -> nn.Module:
         return self.embeddings.patch_embed
 
-    @deprecate_kwarg("attention_mask", version="v4.58.0")
     @check_model_inputs(tie_last_hidden_states=False)
     @auto_docstring
     def forward(
         self,
         pixel_values,
-        attention_mask: Optional[torch.Tensor] = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> BaseModelOutputWithPooling:
         r"""

src/transformers/models/altclip/processing_altclip.py

@@ -17,7 +17,6 @@
 """
 
 from ...processing_utils import ProcessorMixin
-from ...utils.deprecation import deprecate_kwarg
 
 
 class AltCLIPProcessor(ProcessorMixin):
@@ -39,7 +38,6 @@ class AltCLIPProcessor(ProcessorMixin):
     image_processor_class = ("CLIPImageProcessor", "CLIPImageProcessorFast")
     tokenizer_class = ("XLMRobertaTokenizer", "XLMRobertaTokenizerFast")
 
-    @deprecate_kwarg(old_name="feature_extractor", version="5.0.0", new_name="image_processor")
     def __init__(self, image_processor=None, tokenizer=None):
         super().__init__(image_processor, tokenizer)
 

src/transformers/models/beit/modeling_beit.py

@@ -16,7 +16,6 @@
 
 import collections.abc
 import math
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Union
 
@@ -163,14 +162,7 @@ def forward(
         self,
         pixel_values: torch.Tensor,
         bool_masked_pos: Optional[torch.BoolTensor] = None,
-        interpolate_pos_encoding: Optional[bool] = None,
     ) -> torch.Tensor:
-        if self.position_embeddings is not None and interpolate_pos_encoding is not None:
-            warnings.warn(
-                "`interpolate_pos_encoding` argument has no effect for BEiTEmbeddings, embeddings are always "
-                "interpolated to the input image size. The argument will be removed in transformers v4.51.0."
-            )
-
         _, _, height, width = pixel_values.shape
         embeddings, (patch_height, patch_width) = self.patch_embeddings(pixel_values)
         batch_size, seq_len, _ = embeddings.size()
@@ -323,21 +315,6 @@ def forward(
         interpolate_pos_encoding: bool = False,
         resolution: Optional[tuple[int]] = None,
     ) -> Union[tuple[torch.Tensor], tuple[torch.Tensor, torch.Tensor]]:
-        if output_attentions:
-            logger.warning_once(
-                "`BeitSdpaSelfAttention` is used but `torch.nn.functional.scaled_dot_product_attention` does not "
-                "support `output_attentions=True`. Falling back to the manual attention implementation, "
-                "but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. "
-                'This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
-            )
-            return super().forward(
-                hidden_states=hidden_states,
-                output_attentions=output_attentions,
-                relative_position_bias=relative_position_bias,
-                interpolate_pos_encoding=interpolate_pos_encoding,
-                resolution=resolution,
-            )
-
         batch_size, seq_length, _ = hidden_states.shape
         query_layer = (
             self.query(hidden_states)

src/transformers/models/blip_2/modeling_blip_2.py

@@ -15,7 +15,6 @@
 """PyTorch BLIP-2 model."""
 
 import math
-import warnings
 from collections.abc import Callable
 from dataclasses import dataclass
 from typing import Any, Optional, Union
@@ -1088,7 +1087,6 @@ def get_text_features(
         decoder_input_ids: Optional[torch.Tensor] = None,
         decoder_attention_mask: Optional[torch.Tensor] = None,
         labels: Optional[torch.Tensor] = None,
-        legacy_output: bool = True,
     ) -> Union[torch.FloatTensor, CausalLMOutputWithPast]:
         r"""
         decoder_input_ids (`torch.LongTensor` of shape `(batch_size, target_sequence_length)`, *optional*):
@@ -1127,13 +1125,6 @@ def get_text_features(
         ...     text_features = model.get_text_features(**inputs)
         ```"""
 
-        if legacy_output:
-            warnings.warn(
-                "Deprecation notice: In Transformers v4.59, the default return value of `get_text_features` will change. "
-                "Currently, this method returns a model output object, but starting in v4.59, it will return a tensor instead. "
-                "To opt in to the new behavior now, set `legacy_output=False`."
-            )
-
         if self.config.use_decoder_only_language_model:
             text_outputs: CausalLMOutputWithPast = self.language_model(
                 input_ids=input_ids,
@@ -1151,15 +1142,14 @@ def get_text_features(
                 return_dict=True,
             )
 
-        return text_outputs if legacy_output else text_outputs.logits
+        return text_outputs.logits
 
     @filter_out_non_signature_kwargs()
     @auto_docstring
     def get_image_features(
         self,
         pixel_values: torch.FloatTensor,
         interpolate_pos_encoding: bool = False,
-        legacy_output: bool = True,
     ) -> Union[torch.FloatTensor, CausalLMOutputWithPast]:
         r"""
         legacy_output (`bool`, *optional*, defaults to `True`):
@@ -1185,28 +1175,20 @@ def get_image_features(
         >>> with torch.inference_mode():
         ...     image_outputs = model.get_image_features(**inputs)
         ```"""
-        if legacy_output:
-            warnings.warn(
-                "Deprecation notice: In Transformers v4.59, the default return value of `get_text_features` will change. "
-                "Currently, this method returns a model output object, but starting in v4.59, it will return a tensor instead. "
-                "To opt in to the new behavior now, set `legacy_output=False`."
-            )
-
         vision_outputs = self.vision_model(
             pixel_values=pixel_values,
             interpolate_pos_encoding=interpolate_pos_encoding,
             return_dict=True,
         )
 
-        return vision_outputs if legacy_output else vision_outputs.pooler_output
+        return vision_outputs.pooler_output
 
     @filter_out_non_signature_kwargs()
     @auto_docstring
     def get_qformer_features(
         self,
         pixel_values: torch.FloatTensor,
         interpolate_pos_encoding: bool = False,
-        legacy_output: bool = True,
     ) -> Union[torch.FloatTensor, BaseModelOutputWithPooling]:
         r"""
         legacy_output (`bool`, *optional*, defaults to `True`):
@@ -1233,14 +1215,6 @@ def get_qformer_features(
         >>> with torch.inference_mode():
         ...     qformer_outputs = model.get_qformer_features(**inputs)
         ```"""
-
-        if legacy_output:
-            warnings.warn(
-                "Deprecation notice: In Transformers v4.59, the default return value of `get_qformer_features` will change. "
-                "Currently, this method returns a model output object, but starting in v4.59, it will return a tensor instead. "
-                "To opt in to the new behavior now, set `legacy_output=False`."
-            )
-
         vision_outputs: BaseModelOutputWithPooling = self.vision_model(
             pixel_values=pixel_values,
             interpolate_pos_encoding=interpolate_pos_encoding,
@@ -1260,7 +1234,7 @@ def get_qformer_features(
             return_dict=True,
         )
 
-        return query_outputs if legacy_output else query_outputs.last_hidden_state
+        return query_outputs.last_hidden_state
 
     def get_placeholder_mask(self, input_ids: torch.LongTensor, inputs_embeds: torch.FloatTensor):
         """

src/transformers/models/bloom/modeling_bloom.py

@@ -15,7 +15,6 @@
 """PyTorch BLOOM model."""
 
 import math
-import warnings
 from typing import Optional, Union
 
 import torch
@@ -491,7 +490,6 @@ def forward(
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
-        **deprecated_arguments,
     ) -> Union[tuple[torch.Tensor, ...], BaseModelOutputWithPastAndCrossAttentions]:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
@@ -506,16 +504,6 @@ def forward(
 
             [What are input IDs?](../glossary#input-ids)
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
-
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         output_hidden_states = (
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
@@ -823,7 +811,7 @@ def forward(
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
-        **deprecated_arguments,
+        **kwargs,
     ) -> Union[tuple[torch.Tensor], CausalLMOutputWithCrossAttentions]:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
@@ -842,18 +830,6 @@ def forward(
             `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
             are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
         """
-        # Bloom has deprecated kwargs, so we need to pop num_items_in_batch explicitly
-        num_items_in_batch = deprecated_arguments.pop("num_items_in_batch", None)
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
-
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(
@@ -880,7 +856,7 @@ def forward(
                 lm_logits,
                 labels,
                 vocab_size=self.config.vocab_size,
-                num_items_in_batch=num_items_in_batch,
+                num_items_in_batch=kwargs["num_items_in_batch"],
             )
 
         if not return_dict:
@@ -932,7 +908,6 @@ def forward(
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[tuple[torch.Tensor], SequenceClassifierOutputWithPast]:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
@@ -951,16 +926,6 @@ def forward(
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
-
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(
@@ -1066,7 +1031,6 @@ def forward(
         output_attentions: Optional[bool] = None,
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
-        **deprecated_arguments,
     ) -> Union[tuple[torch.Tensor], TokenClassifierOutput]:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
@@ -1085,16 +1049,6 @@ def forward(
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
-        if deprecated_arguments.pop("position_ids", False) is not False:
-            # `position_ids` could have been `torch.Tensor` or `None` so defaulting pop to `False` allows to detect if users were passing explicitly `None`
-            warnings.warn(
-                "`position_ids` have no functionality in BLOOM and will be removed in v5.0.0. You can safely ignore"
-                " passing `position_ids`.",
-                FutureWarning,
-            )
-        if len(deprecated_arguments) > 0:
-            raise ValueError(f"Got unexpected arguments: {deprecated_arguments}")
-
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(

src/transformers/models/clap/processing_clap.py

@@ -16,13 +16,8 @@
 Audio/Text processor class for CLAP
 """
 
-from typing import Optional, Union
-
-from ...audio_utils import AudioInput
-from ...processing_utils import ProcessingKwargs, ProcessorMixin, Unpack
-from ...tokenization_utils_base import PreTokenizedInput, TextInput
+from ...processing_utils import ProcessorMixin
 from ...utils import logging
-from ...utils.deprecation import deprecate_kwarg
 
 
 logger = logging.get_logger(__name__)
@@ -48,28 +43,5 @@ class ClapProcessor(ProcessorMixin):
     def __init__(self, feature_extractor, tokenizer):
         super().__init__(feature_extractor, tokenizer)
 
-    @deprecate_kwarg("audios", version="v4.59.0", new_name="audio")
-    def __call__(
-        self,
-        text: Optional[Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]]] = None,
-        audios: Optional[AudioInput] = None,
-        audio: Optional[AudioInput] = None,
-        **kwargs: Unpack[ProcessingKwargs],
-    ):
-        """
-        Forwards the `audio` and `sampling_rate` arguments to [`~ClapFeatureExtractor.__call__`] and the `text`
-        argument to [`~RobertaTokenizerFast.__call__`]. Please refer to the docstring of the above two methods for more
-        information.
-        """
-        # The `deprecate_kwarg` will not work if the inputs are passed as arguments, so we check
-        # again that the correct naming is used
-        if audios is not None and audio is None:
-            logger.warning(
-                "Using `audios` keyword argument is deprecated when calling ClapProcessor, instead use `audio`."
-            )
-            audio = audios
-
-        return super().__call__(text=text, audio=audio, **kwargs)
-
 
 __all__ = ["ClapProcessor"]