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

transformers/pipelines/automatic_speech_recognition.py

@@ -480,10 +480,8 @@ class AutomaticSpeechRecognitionPipeline(ChunkPipeline):
                         inputs,
                         sampling_rate=self.feature_extractor.sampling_rate,
                         return_tensors="pt",
-                        return_token_timestamps=True,
                         return_attention_mask=True,
                     )
-                    extra["num_frames"] = processed.pop("num_frames")
                 else:
                     processed = self.feature_extractor(
                         inputs,

transformers/pipelines/base.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2018 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -12,6 +11,8 @@
 # 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 __future__ import annotations
+
 import collections
 import copy
 import csv
@@ -26,13 +27,12 @@ from abc import ABC, abstractmethod
 from collections import UserDict
 from contextlib import contextmanager
 from os.path import abspath, exists
-from typing import TYPE_CHECKING, Any, Optional, Union
+from typing import TYPE_CHECKING, Any, Union
 
 from ..dynamic_module_utils import custom_object_save
 from ..feature_extraction_utils import PreTrainedFeatureExtractor
 from ..generation import GenerationConfig
 from ..image_processing_utils import BaseImageProcessor
-from ..modelcard import ModelCard
 from ..models.auto import AutoConfig, AutoTokenizer
 from ..processing_utils import ProcessorMixin
 from ..tokenization_python import PreTrainedTokenizer
@@ -306,10 +306,10 @@ def get_default_model_and_revision(targeted_task: dict, task_options: Any | None
 
 
 def load_assistant_model(
-    model: "PreTrainedModel",
-    assistant_model: Union[str, "PreTrainedModel"] | None,
+    model: PreTrainedModel,
+    assistant_model: str | PreTrainedModel | None,
     assistant_tokenizer: PreTrainedTokenizer | None,
-) -> tuple[Optional["PreTrainedModel"], PreTrainedTokenizer | None]:
+) -> tuple[PreTrainedModel | None, PreTrainedTokenizer | None]:
     """
     Prepares the assistant model and the assistant tokenizer for a pipeline whose model that can call `generate`.
 
@@ -466,7 +466,7 @@ class PipelineDataFormat:
         input_path: str | None,
         column: str | None,
         overwrite=False,
-    ) -> "PipelineDataFormat":
+    ) -> PipelineDataFormat:
         """
         Creates an instance of the right subclass of [`~pipelines.PipelineDataFormat`] depending on `format`.
 
@@ -677,8 +677,6 @@ def build_pipeline_init_args(
             [`ProcessorMixin`]. Processor is a composite object that might contain `tokenizer`, `feature_extractor`, and
             `image_processor`."""
     docstring += r"""
-        modelcard (`str` or [`ModelCard`], *optional*):
-            Model card attributed to the model for this pipeline.
         task (`str`, defaults to `""`):
             A task-identifier for the pipeline.
         num_workers (`int`, *optional*, defaults to 8):
@@ -777,14 +775,13 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
 
     def __init__(
         self,
-        model: "PreTrainedModel",
+        model: PreTrainedModel,
         tokenizer: PreTrainedTokenizer | None = None,
-        feature_extractor: Optional[PreTrainedFeatureExtractor] = None,
+        feature_extractor: PreTrainedFeatureExtractor | None = None,
         image_processor: BaseImageProcessor | None = None,
         processor: ProcessorMixin | None = None,
-        modelcard: ModelCard | None = None,
         task: str = "",
-        device: Union[int, "torch.device"] | None = None,
+        device: int | torch.device | None = None,
         binary_output: bool = False,
         **kwargs,
     ):
@@ -797,7 +794,6 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
         self.feature_extractor = feature_extractor
         self.image_processor = image_processor
         self.processor = processor
-        self.modelcard = modelcard
 
         # `accelerate` device map
         hf_device_map = getattr(self.model, "hf_device_map", None)
@@ -905,7 +901,7 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
             # Update the generation config with task specific params if they exist.
             # NOTE: 1. `prefix` is pipeline-specific and doesn't exist in the generation config.
             #       2. `task_specific_params` is a legacy feature and should be removed in a future version.
-            task_specific_params = self.model.config.task_specific_params
+            task_specific_params = getattr(self.model.config, "task_specific_params", None)
             if task_specific_params is not None and task in task_specific_params:
                 this_task_params = task_specific_params.get(task)
                 if "prefix" in this_task_params:
@@ -1012,14 +1008,14 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
         return self(X)
 
     @property
-    def dtype(self) -> Optional["torch.dtype"]:
+    def dtype(self) -> torch.dtype | None:
         """
         Dtype of the model (if it's Pytorch model), `None` otherwise.
         """
         return getattr(self.model, "dtype", None)
 
     @property
-    def torch_dtype(self) -> Optional["torch.dtype"]:
+    def torch_dtype(self) -> torch.dtype | None:
         """
         Torch dtype of the model (if it's Pytorch model), `None` otherwise.
         """

transformers/pipelines/image_text_to_text.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -335,7 +334,7 @@ class ImageTextToTextPipeline(Pipeline):
                 return_tensors="pt",
                 tokenize=True,
                 return_dict=True,
-            )
+            ).to(dtype=self.dtype)
             model_inputs["text"] = inputs
             return model_inputs
         # In case we only have text inputs

transformers/pipelines/question_answering.py

@@ -1,13 +1,11 @@
 import inspect
 import types
-import warnings
 from collections.abc import Iterable
 from typing import TYPE_CHECKING
 
 import numpy as np
 
 from ..data import SquadExample, SquadFeatures, squad_convert_examples_to_features
-from ..modelcard import ModelCard
 from ..tokenization_python import PreTrainedTokenizer
 from ..utils import (
     PaddingStrategy,
@@ -179,18 +177,6 @@ class QuestionAnsweringArgumentHandler(ArgumentHandler):
                 inputs = list(args)
         # Generic compatibility with sklearn and Keras
         # Batched data
-        elif "X" in kwargs:
-            warnings.warn(
-                "Passing the `X` argument to the pipeline is deprecated and will be removed in v5. Inputs should be passed using the `question` and `context` keyword arguments instead.",
-                FutureWarning,
-            )
-            inputs = kwargs["X"]
-        elif "data" in kwargs:
-            warnings.warn(
-                "Passing the `data` argument to the pipeline is deprecated and will be removed in v5. Inputs should be passed using the `question` and `context` keyword arguments instead.",
-                FutureWarning,
-            )
-            inputs = kwargs["data"]
         elif "question" in kwargs and "context" in kwargs:
             if isinstance(kwargs["question"], list) and isinstance(kwargs["context"], str):
                 inputs = [{"question": Q, "context": kwargs["context"]} for Q in kwargs["question"]]
@@ -255,22 +241,8 @@ class QuestionAnsweringPipeline(ChunkPipeline):
     default_input_names = "question,context"
     handle_impossible_answer = False
 
-    def __init__(
-        self,
-        model: "PreTrainedModel",
-        tokenizer: PreTrainedTokenizer,
-        modelcard: ModelCard | None = None,
-        task: str = "",
-        **kwargs,
-    ):
-        super().__init__(
-            model=model,
-            tokenizer=tokenizer,
-            modelcard=modelcard,
-            task=task,
-            **kwargs,
-        )
-
+    def __init__(self, model: "PreTrainedModel", tokenizer: PreTrainedTokenizer, task: str = "", **kwargs):
+        super().__init__(model=model, tokenizer=tokenizer, task=task, **kwargs)
         self._args_parser = QuestionAnsweringArgumentHandler()
         self.check_model_type(MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES)
 
@@ -297,7 +269,6 @@ class QuestionAnsweringPipeline(ChunkPipeline):
     def _sanitize_parameters(
         self,
         padding=None,
-        topk=None,
         top_k=None,
         doc_stride=None,
         max_answer_len=None,
@@ -319,9 +290,6 @@ class QuestionAnsweringPipeline(ChunkPipeline):
             preprocess_params["max_seq_len"] = max_seq_len
 
         postprocess_params = {}
-        if topk is not None and top_k is None:
-            warnings.warn("topk parameter is deprecated, use top_k instead", UserWarning)
-            top_k = topk
         if top_k is not None:
             if top_k < 1:
                 raise ValueError(f"top_k parameter should be >= 1 (got {top_k})")
@@ -336,7 +304,7 @@ class QuestionAnsweringPipeline(ChunkPipeline):
             postprocess_params["align_to_words"] = align_to_words
         return preprocess_params, {}, postprocess_params
 
-    def __call__(self, *args, **kwargs):
+    def __call__(self, **kwargs):
         """
         Answer the question(s) given as inputs by using the context(s).
 
@@ -374,14 +342,7 @@ class QuestionAnsweringPipeline(ChunkPipeline):
             - **answer** (`str`) -- The answer to the question.
         """
 
-        # Convert inputs to features
-        if args:
-            warnings.warn(
-                "Passing a list of SQuAD examples to the pipeline is deprecated and will be removed in v5. Inputs should be passed using the `question` and `context` keyword arguments instead.",
-                FutureWarning,
-            )
-
-        examples = self._args_parser(*args, **kwargs)
+        examples = self._args_parser(**kwargs)
         if isinstance(examples, (list, tuple)) and len(examples) == 1:
             return super().__call__(examples[0], **kwargs)
         return super().__call__(examples, **kwargs)

transformers/pipelines/text_classification.py

@@ -1,5 +1,4 @@
 import inspect
-import warnings
 from typing import Any
 
 import numpy as np
@@ -88,28 +87,16 @@ class TextClassificationPipeline(Pipeline):
 
         self.check_model_type(MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES)
 
-    def _sanitize_parameters(self, return_all_scores=None, function_to_apply=None, top_k="", **tokenizer_kwargs):
+    def _sanitize_parameters(self, function_to_apply=None, top_k="", **tokenizer_kwargs):
         # Using "" as default argument because we're going to use `top_k=None` in user code to declare
         # "No top_k"
         preprocess_params = tokenizer_kwargs
 
         postprocess_params = {}
-        if hasattr(self.model.config, "return_all_scores") and return_all_scores is None:
-            return_all_scores = self.model.config.return_all_scores
 
         if isinstance(top_k, int) or top_k is None:
             postprocess_params["top_k"] = top_k
             postprocess_params["_legacy"] = False
-        elif return_all_scores is not None:
-            warnings.warn(
-                "`return_all_scores` is now deprecated,  if want a similar functionality use `top_k=None` instead of"
-                " `return_all_scores=True` or `top_k=1` instead of `return_all_scores=False`.",
-                UserWarning,
-            )
-            if return_all_scores:
-                postprocess_params["top_k"] = None
-            else:
-                postprocess_params["top_k"] = 1
 
         if isinstance(function_to_apply, str):
             function_to_apply = ClassificationFunction[function_to_apply.upper()]

transformers/pipelines/text_to_audio.py

@@ -152,8 +152,12 @@ class TextToAudioPipeline(Pipeline):
 
         if self.model.config.model_type == "bark":
             # bark Tokenizer is called with BarkProcessor which uses those kwargs
+            # Check if generation_config has semantic_config (BarkGenerationConfig) or use default
+            max_length = 256
+            if hasattr(self.generation_config, "semantic_config"):
+                max_length = getattr(self.generation_config.semantic_config, "max_input_semantic_length", 256)
             new_kwargs = {
-                "max_length": self.generation_config.semantic_config.get("max_input_semantic_length", 256),
+                "max_length": max_length,
                 "add_special_tokens": False,
                 "return_attention_mask": True,
                 "return_token_type_ids": False,

transformers/pipelines/token_classification.py

@@ -147,8 +147,6 @@ class TokenClassificationPipeline(ChunkPipeline):
     def _sanitize_parameters(
         self,
         ignore_labels=None,
-        grouped_entities: bool | None = None,
-        ignore_subwords: bool | None = None,
         aggregation_strategy: AggregationStrategy | None = None,
         offset_mapping: list[tuple[int, int]] | None = None,
         is_split_into_words: bool = False,
@@ -165,25 +163,6 @@ class TokenClassificationPipeline(ChunkPipeline):
             preprocess_params["offset_mapping"] = offset_mapping
 
         postprocess_params = {}
-        if grouped_entities is not None or ignore_subwords is not None:
-            if grouped_entities and ignore_subwords:
-                aggregation_strategy = AggregationStrategy.FIRST
-            elif grouped_entities and not ignore_subwords:
-                aggregation_strategy = AggregationStrategy.SIMPLE
-            else:
-                aggregation_strategy = AggregationStrategy.NONE
-
-            if grouped_entities is not None:
-                warnings.warn(
-                    "`grouped_entities` is deprecated and will be removed in version v5.0.0, defaulted to"
-                    f' `aggregation_strategy="{aggregation_strategy}"` instead.'
-                )
-            if ignore_subwords is not None:
-                warnings.warn(
-                    "`ignore_subwords` is deprecated and will be removed in version v5.0.0, defaulted to"
-                    f' `aggregation_strategy="{aggregation_strategy}"` instead.'
-                )
-
         if aggregation_strategy is not None:
             if isinstance(aggregation_strategy, str):
                 aggregation_strategy = AggregationStrategy[aggregation_strategy.upper()]
@@ -279,7 +258,7 @@ class TokenClassificationPipeline(ChunkPipeline):
                 raise ValueError("When `is_split_into_words=True`, `sentence` must be a list of tokens.")
             words = sentence
             sentence = delimiter.join(words)  # Recreate the sentence string for later display and slicing
-            # This map will allows to convert back word => char indices
+            # This map will allow to convert back word => char indices
             word_to_chars_map = []
             delimiter_len = len(delimiter)
             char_offset = 0

transformers/pipelines/video_classification.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.
-import warnings
 from io import BytesIO
 from typing import Any, overload
 
@@ -88,7 +87,7 @@ class VideoClassificationPipeline(Pipeline):
     @overload
     def __call__(self, inputs: list[str], **kwargs: Any) -> list[list[dict[str, Any]]]: ...
 
-    def __call__(self, inputs: str | list[str] | None = None, **kwargs):
+    def __call__(self, inputs: str | list[str] | None, **kwargs):
         """
         Assign labels to the video(s) passed as inputs.
 
@@ -126,13 +125,6 @@ class VideoClassificationPipeline(Pipeline):
             - **label** (`str`) -- The label identified by the model.
             - **score** (`int`) -- The score attributed by the model for that label.
         """
-        # After deprecation of this is completed, remove the default `None` value for `images`
-        if "videos" in kwargs:
-            warnings.warn(
-                "The `videos` argument has been renamed to `inputs`. In version 5 of Transformers, `videos` will no longer be accepted",
-                FutureWarning,
-            )
-            inputs = kwargs.pop("videos")
         if inputs is None:
             raise ValueError("Cannot call the video-classification pipeline without an inputs argument!")
         return super().__call__(inputs, **kwargs)

transformers/pipelines/zero_shot_audio_classification.py

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

transformers/pipelines/zero_shot_classification.py

@@ -145,12 +145,6 @@ class ZeroShotClassificationPipeline(ChunkPipeline):
         return inputs
 
     def _sanitize_parameters(self, **kwargs):
-        if kwargs.get("multi_class") is not None:
-            kwargs["multi_label"] = kwargs["multi_class"]
-            logger.warning(
-                "The `multi_class` argument has been deprecated and renamed to `multi_label`. "
-                "`multi_class` will be removed in a future version of Transformers."
-            )
         preprocess_params = {}
         if "candidate_labels" in kwargs:
             preprocess_params["candidate_labels"] = self._args_parser._parse_labels(kwargs["candidate_labels"])

transformers/pipelines/zero_shot_image_classification.py

@@ -1,4 +1,3 @@
-import warnings
 from collections import UserDict
 from typing import Any, Union, overload
 
@@ -132,12 +131,6 @@ class ZeroShotImageClassificationPipeline(Pipeline):
             preprocess_params["timeout"] = kwargs["timeout"]
         if "hypothesis_template" in kwargs:
             preprocess_params["hypothesis_template"] = kwargs["hypothesis_template"]
-        if tokenizer_kwargs is not None:
-            warnings.warn(
-                "The `tokenizer_kwargs` argument is deprecated and will be removed in version 5 of Transformers",
-                FutureWarning,
-            )
-            preprocess_params["tokenizer_kwargs"] = tokenizer_kwargs
 
         return preprocess_params, {}, {}