Remove deprecated (#8604)

* Remove old deprecated arguments

Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>

* Remove needless imports

* Fix tests

Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>

examples/seq2seq/test_finetune_trainer.py

@@ -138,7 +138,7 @@ class TestFinetuneTrainer(TestCasePlus):
             per_device_train_batch_size=batch_size,
             per_device_eval_batch_size=batch_size,
             predict_with_generate=True,
-            evaluate_during_training=True,
+            evaluation_strategy="steps",
             do_train=True,
             do_eval=True,
             warmup_steps=0,
@@ -179,7 +179,7 @@ class TestFinetuneTrainer(TestCasePlus):
             --per_device_eval_batch_size 4
             --learning_rate 3e-3
             --warmup_steps 8
-            --evaluate_during_training
+            --evaluation_strategy steps
             --predict_with_generate
             --logging_steps 0
             --save_steps {str(eval_steps)}

examples/token-classification/run_ner_old.py

@@ -254,7 +254,7 @@ def main():
         trainer.save_model()
         # For convenience, we also re-save the tokenizer to the same directory,
         # so that you can share your model easily on huggingface.co/models =)
-        if trainer.is_world_master():
+        if trainer.is_world_process_zero():
             tokenizer.save_pretrained(training_args.output_dir)
 
     # Evaluation
@@ -265,7 +265,7 @@ def main():
         result = trainer.evaluate()
 
         output_eval_file = os.path.join(training_args.output_dir, "eval_results.txt")
-        if trainer.is_world_master():
+        if trainer.is_world_process_zero():
             with open(output_eval_file, "w") as writer:
                 logger.info("***** Eval results *****")
                 for key, value in result.items():

src/transformers/data/processors/squad.py

@@ -145,11 +145,11 @@ def squad_convert_example_to_features(
     # in the way they compute mask of added tokens.
     tokenizer_type = type(tokenizer).__name__.replace("Tokenizer", "").lower()
     sequence_added_tokens = (
-        tokenizer.max_len - tokenizer.max_len_single_sentence + 1
+        tokenizer.model_max_length - tokenizer.max_len_single_sentence + 1
         if tokenizer_type in MULTI_SEP_TOKENS_TOKENIZERS_SET
-        else tokenizer.max_len - tokenizer.max_len_single_sentence
+        else tokenizer.model_max_length - tokenizer.max_len_single_sentence
     )
-    sequence_pair_added_tokens = tokenizer.max_len - tokenizer.max_len_sentences_pair
+    sequence_pair_added_tokens = tokenizer.model_max_length - tokenizer.max_len_sentences_pair
 
     span_doc_tokens = all_doc_tokens
     while len(spans) * doc_stride < len(all_doc_tokens):

src/transformers/models/albert/modeling_albert.py

@@ -16,7 +16,6 @@
 
 import math
 import os
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -742,7 +741,6 @@ class AlbertForPreTraining(AlbertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
         r"""
         labels (``torch.LongTensor`` of shape ``(batch_size, sequence_length)``, `optional`):
@@ -753,8 +751,6 @@ class AlbertForPreTraining(AlbertPreTrainedModel):
             Labels for computing the next sequence prediction (classification) loss. Input should be a sequence pair
             (see :obj:`input_ids` docstring) Indices should be in ``[0, 1]``. ``0`` indicates original order (sequence
             A, then sequence B), ``1`` indicates switched order (sequence B, then sequence A).
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
 
         Returns:
 
@@ -773,14 +769,6 @@ class AlbertForPreTraining(AlbertPreTrainedModel):
             >>> sop_logits = outputs.sop_logits
 
         """
-
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.albert(
@@ -898,23 +886,13 @@ class AlbertForMaskedLM(AlbertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
             Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,
             config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored
             (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.albert(

src/transformers/models/bart/modeling_bart.py

@@ -15,7 +15,6 @@
 """PyTorch BART model, ported from the fairseq repo."""
 import math
 import random
-import warnings
 from typing import Dict, List, Optional, Tuple
 
 import numpy as np
@@ -529,7 +528,6 @@ class BartDecoder(nn.Module):
         output_attentions=False,
         output_hidden_states=False,
         return_dict=True,
-        **unused,
     ):
         """
         Includes several features from "Jointly Learning to Align and Translate with Transformer Models" (Garg et al.,
@@ -551,18 +549,6 @@ class BartDecoder(nn.Module):
                 - hidden states
                 - attentions
         """
-        if "decoder_cached_states" in unused:
-            warnings.warn(
-                "The `decoder_cached_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_cached_states")
-        if "decoder_past_key_values" in unused:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_past_key_values")
 
         # check attention mask and invert
         if encoder_padding_mask is not None:
@@ -873,14 +859,7 @@ class BartModel(PretrainedBartModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
-        if "decoder_past_key_values" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_values")
 
         if decoder_input_ids is None:
             use_cache = False
@@ -1006,7 +985,6 @@ class BartForConditionalGeneration(PretrainedBartModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **unused,
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -1034,24 +1012,6 @@ class BartForConditionalGeneration(PretrainedBartModel):
             >>> tokenizer.decode(predictions).split()
             >>> # ['good', 'great', 'all', 'really', 'very']
         """
-        if "lm_labels" in unused:
-            warnings.warn(
-                "The `lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = unused.pop("lm_labels")
-        if "decoder_cached_states" in unused:
-            warnings.warn(
-                "The `decoder_cached_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_cached_states")
-        if "decoder_past_key_values" in unused:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_past_key_values")
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         if labels is not None:

src/transformers/models/bert/modeling_bert.py

@@ -896,7 +896,6 @@ class BertForPreTraining(BertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape ``(batch_size, sequence_length)``, `optional`):
@@ -928,13 +927,6 @@ class BertForPreTraining(BertPreTrainedModel):
             >>> prediction_logits = outputs.prediction_logits
             >>> seq_relationship_logits = outputs.seq_relationship_logits
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.bert(
@@ -1136,24 +1128,13 @@ class BertForMaskedLM(BertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
             Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,
             config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored
             (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert "lm_labels" not in kwargs, "Use `BertWithLMHead` for autoregressive language modeling task."
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 

src/transformers/models/ctrl/modeling_ctrl.py

@@ -15,9 +15,6 @@
 # limitations under the License.
 """ PyTorch CTRL model."""
 
-
-import warnings
-
 import numpy as np
 import torch
 import torch.nn as nn
@@ -369,15 +366,7 @@ class CTRLModel(CTRLPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
-        if "past" in kwargs:
-            warnings.warn(
-                "The `past` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("past")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
 
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
         use_cache = use_cache if use_cache is not None else self.config.use_cache
@@ -542,7 +531,6 @@ class CTRLLMHeadModel(CTRLPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -550,13 +538,6 @@ class CTRLLMHeadModel(CTRLPreTrainedModel):
             ``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]``
         """
-        if "past" in kwargs:
-            warnings.warn(
-                "The `past` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("past")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(

src/transformers/models/distilbert/modeling_distilbert.py

@@ -20,7 +20,6 @@
 
 import copy
 import math
-import warnings
 
 import numpy as np
 import torch
@@ -526,23 +525,13 @@ class DistilBertForMaskedLM(DistilBertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
             Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,
             config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored
             (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``.
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         dlbrt_output = self.distilbert(

src/transformers/models/electra/modeling_electra.py

@@ -16,7 +16,6 @@
 
 import math
 import os
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -1000,23 +999,13 @@ class ElectraForMaskedLM(ElectraPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
             Labels for computing the masked language modeling loss. Indices should be in ``[-100, 0, ...,
             config.vocab_size]`` (see ``input_ids`` docstring) Tokens with indices set to ``-100`` are ignored
             (masked), the loss is only computed for the tokens with labels in ``[0, ..., config.vocab_size]``
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         generator_hidden_states = self.electra(

src/transformers/models/fsmt/modeling_fsmt.py

@@ -29,7 +29,6 @@
 
 import math
 import random
-import warnings
 from typing import Any, Dict, List, Optional, Tuple
 
 import torch
@@ -618,7 +617,6 @@ class FSMTDecoder(nn.Module):
         output_attentions=False,
         output_hidden_states=False,
         return_dict=True,
-        **unused,
     ):
         """
         Includes several features from "Jointly Learning to Align and Translate with Transformer Models" (Garg et al.,
@@ -640,19 +638,6 @@ class FSMTDecoder(nn.Module):
                 - hidden states
                 - attentions
         """
-        if "decoder_cached_states" in unused:
-            warnings.warn(
-                "The `decoder_cached_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_cached_states")
-        if "decoder_past_key_values" in unused:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = unused.pop("decoder_past_key_values")
-
         # check attention mask and invert
         if encoder_padding_mask is not None:
             encoder_padding_mask = invert_mask(encoder_padding_mask)
@@ -933,15 +918,7 @@ class FSMTModel(PretrainedFSMTModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
-        if "decoder_past_key_values" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_values")
-
         if decoder_input_ids is None:
             use_cache = False
 
@@ -1071,7 +1048,6 @@ class FSMTForConditionalGeneration(PretrainedFSMTModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **unused,
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):

src/transformers/models/gpt2/modeling_gpt2.py

@@ -16,7 +16,6 @@
 """PyTorch OpenAI GPT-2 model."""
 
 import os
-import warnings
 from dataclasses import dataclass
 from typing import List, Optional, Tuple
 
@@ -528,16 +527,7 @@ class GPT2Model(GPT2PreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
-        if "past" in kwargs:
-            warnings.warn(
-                "The `past` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("past")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
-
         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
@@ -758,7 +748,6 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -766,13 +755,6 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
             ``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]``
         """
-        if "past" in kwargs:
-            warnings.warn(
-                "The `past` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("past")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(
@@ -900,8 +882,6 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
             Labels for computing the multiple choice classification loss. Indices should be in ``[0, ...,
             num_choices]`` where `num_choices` is the size of the second dimension of the input tensors. (see
             `input_ids` above)
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
 
         Return:
 
@@ -930,19 +910,6 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
             >>> mc_logits = outputs.mc_logits
 
         """
-        if "lm_labels" in kwargs:
-            warnings.warn(
-                "The `lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("lm_labels")
-        if "past" in kwargs:
-            warnings.warn(
-                "The `past` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("past")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         transformer_outputs = self.transformer(

src/transformers/models/gpt2/tokenization_gpt2.py

@@ -17,7 +17,6 @@
 
 import json
 import os
-import warnings
 from functools import lru_cache
 from typing import Optional, Tuple
 
@@ -293,13 +292,6 @@ class GPT2Tokenizer(PreTrainedTokenizer):
         return vocab_file, merge_file
 
     def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         add_prefix_space = kwargs.pop("add_prefix_space", self.add_prefix_space)
         if is_split_into_words or add_prefix_space:
             text = " " + text

src/transformers/models/gpt2/tokenization_gpt2_fast.py

@@ -16,7 +16,6 @@
 
 
 import json
-import warnings
 from typing import Optional, Tuple
 
 from tokenizers import pre_tokenizers
@@ -151,13 +150,6 @@ class GPT2TokenizerFast(PreTrainedTokenizerFast):
         self.add_prefix_space = add_prefix_space
 
     def _batch_encode_plus(self, *args, **kwargs) -> BatchEncoding:
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         is_split_into_words = kwargs.get("is_split_into_words", False)
         assert self.add_prefix_space or not is_split_into_words, (
             f"You need to instantiate {self.__class__.__name__} with add_prefix_space=True "
@@ -167,14 +159,7 @@ class GPT2TokenizerFast(PreTrainedTokenizerFast):
         return super()._batch_encode_plus(*args, **kwargs)
 
     def _encode_plus(self, *args, **kwargs) -> BatchEncoding:
-        if "is_pretokenized" in kwargs:
+        is_split_into_words = kwargs.get("is_split_into_words", False)
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-        else:
-            is_split_into_words = kwargs.get("is_split_into_words", False)
 
         assert self.add_prefix_space or not is_split_into_words, (
             f"You need to instantiate {self.__class__.__name__} with add_prefix_space=True "

src/transformers/models/longformer/modeling_longformer.py

@@ -15,7 +15,6 @@
 """PyTorch Longformer model. """
 
 import math
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -1509,7 +1508,6 @@ class LongformerForMaskedLM(LongformerPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -1538,14 +1536,6 @@ class LongformerForMaskedLM(LongformerPreTrainedModel):
             >>> loss = outputs.loss
             >>> prediction_logits = output.logits
         """
-
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.longformer(

src/transformers/models/mobilebert/modeling_mobilebert.py

@@ -1109,7 +1109,6 @@ class MobileBertForMaskedLM(MobileBertPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -1119,12 +1118,6 @@ class MobileBertForMaskedLM(MobileBertPreTrainedModel):
         kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
             Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.mobilebert(

src/transformers/models/openai/modeling_openai.py

@@ -19,7 +19,6 @@
 import json
 import math
 import os
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -645,7 +644,6 @@ class OpenAIGPTDoubleHeadsModel(OpenAIGPTPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         mc_token_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, num_choices)`, `optional`, default to index of the last token of the input):
@@ -659,8 +657,6 @@ class OpenAIGPTDoubleHeadsModel(OpenAIGPTPreTrainedModel):
             Labels for computing the multiple choice classification loss. Indices should be in ``[0, ...,
             num_choices]`` where `num_choices` is the size of the second dimension of the input tensors. (see
             `input_ids` above)
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
 
         Return:
 
@@ -683,13 +679,6 @@ class OpenAIGPTDoubleHeadsModel(OpenAIGPTPreTrainedModel):
             >>> mc_logits = outputs.mc_logits
         """
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-        if "lm_labels" in kwargs:
-            warnings.warn(
-                "The `lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
 
         transformer_outputs = self.transformer(
             input_ids,

src/transformers/models/prophetnet/tokenization_prophetnet.py

@@ -302,7 +302,7 @@ class ProphetNetTokenizer(PreTrainedTokenizer):
         **kwargs,
     ) -> BatchEncoding:
         if max_length is None:
-            max_length = self.max_len
+            max_length = self.model_max_length
         model_inputs = self(
             src_texts,
             add_special_tokens=True,

src/transformers/models/roberta/modeling_roberta.py

@@ -16,7 +16,6 @@
 """PyTorch RoBERTa model. """
 
 import math
-import warnings
 
 import torch
 import torch.nn as nn
@@ -872,7 +871,6 @@ class RobertaForMaskedLM(RobertaPreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`):
@@ -882,13 +880,6 @@ class RobertaForMaskedLM(RobertaPreTrainedModel):
         kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
             Used to hide legacy arguments that have been deprecated.
         """
-        if "masked_lm_labels" in kwargs:
-            warnings.warn(
-                "The `masked_lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("masked_lm_labels")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         outputs = self.roberta(

src/transformers/models/roberta/tokenization_roberta.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """Tokenization classes for RoBERTa."""
 
-import warnings
 from typing import List, Optional
 
 from ...tokenization_utils import AddedToken
@@ -251,13 +250,6 @@ class RobertaTokenizer(GPT2Tokenizer):
         return len(cls + token_ids_0 + sep + sep + token_ids_1 + sep) * [0]
 
     def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         add_prefix_space = kwargs.pop("add_prefix_space", self.add_prefix_space)
         if (is_split_into_words or add_prefix_space) and (len(text) > 0 and not text[0].isspace()):
             text = " " + text

src/transformers/models/t5/modeling_t5.py

@@ -18,7 +18,6 @@
 import copy
 import math
 import os
-import warnings
 
 import torch
 import torch.nn.functional as F
@@ -1048,7 +1047,6 @@ class T5Model(T5PreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
         r"""
         Returns:
@@ -1066,20 +1064,6 @@ class T5Model(T5PreTrainedModel):
 
             >>> last_hidden_states = outputs.last_hidden_state
         """
-        if "decoder_past_key_value_states" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_value_states")
-        if "decoder_past_key_values" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_values")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
-
         use_cache = use_cache if use_cache is not None else self.config.use_cache
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
@@ -1198,15 +1182,12 @@ class T5ForConditionalGeneration(T5PreTrainedModel):
         output_attentions=None,
         output_hidden_states=None,
         return_dict=None,
-        **kwargs,
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
             Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[-100, 0, ...,
             config.vocab_size - 1]`. All labels set to ``-100`` are ignored (masked), the loss is only computed for
             labels in ``[0, ..., config.vocab_size]``
-        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
-            Used to hide legacy arguments that have been deprecated.
 
         Returns:
 
@@ -1226,27 +1207,6 @@ class T5ForConditionalGeneration(T5PreTrainedModel):
             >>> input_ids = tokenizer("summarize: studies have shown that owning a dog is good for you ", return_tensors="pt").input_ids  # Batch size 1
             >>> outputs = model.generate(input_ids)
         """
-
-        if "lm_labels" in kwargs:
-            warnings.warn(
-                "The `lm_labels` argument is deprecated and will be removed in a future version, use `labels` instead.",
-                FutureWarning,
-            )
-            labels = kwargs.pop("lm_labels")
-        if "decoder_past_key_value_states" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_value_states")
-        if "decoder_past_key_values" in kwargs:
-            warnings.warn(
-                "The `decoder_past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                FutureWarning,
-            )
-            past_key_values = kwargs.pop("decoder_past_key_values")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
-
         use_cache = use_cache if use_cache is not None else self.config.use_cache
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 

src/transformers/models/t5/modeling_tf_t5.py

@@ -595,7 +595,6 @@ class TFT5MainLayer(tf.keras.layers.Layer):
         output_attentions=None,
         output_hidden_states=None,
         training=False,
-        **kwargs,
     ) -> Tuple:
         if isinstance(inputs, (tuple, list)):
             input_ids = inputs[0]
@@ -621,21 +620,8 @@ class TFT5MainLayer(tf.keras.layers.Layer):
             output_attentions = inputs.get("output_attentions", output_attentions)
             output_hidden_states = inputs.get("output_hidden_states", output_hidden_states)
             assert len(inputs) <= 10, "Too many inputs."
-
-            if "past_key_values" in inputs:
-                warnings.warn(
-                    "The `past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = inputs.pop("past_key_values")
         else:
             input_ids = inputs
-            if "past_key_values" in kwargs:
-                warnings.warn(
-                    "The `past_key_values` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = kwargs.pop("past_key_values")
 
         output_attentions = output_attentions if output_attentions is not None else self.output_attentions
         output_hidden_states = output_hidden_states if output_hidden_states is not None else self.output_hidden_states
@@ -1078,23 +1064,9 @@ class TFT5Model(TFT5PreTrainedModel):
             output_attentions = inputs.get("output_attentions", output_attentions)
             output_hidden_states = inputs.get("output_hidden_states", output_hidden_states)
             assert len(inputs) <= 13, "Too many inputs."
-
-            if "past_key_value_states" in inputs:
-                warnings.warn(
-                    "The `past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = inputs.pop("past_key_value_states")
         else:
             input_ids = inputs
 
-            if "past_key_value_states" in kwargs:
-                warnings.warn(
-                    "The `past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = kwargs.pop("past_key_value_states")
-
         use_cache = use_cache if use_cache is not None else self.config.use_cache
         output_attentions = output_attentions if output_attentions else self.config.output_attentions
         output_hidden_states = output_hidden_states if output_hidden_states else self.config.output_hidden_states
@@ -1294,23 +1266,9 @@ class TFT5ForConditionalGeneration(TFT5PreTrainedModel, TFCausalLanguageModeling
             output_hidden_states = inputs.get("output_hidden_states", output_hidden_states)
             return_dict = inputs.get("return_dict", return_dict)
             assert len(inputs) <= 14, "Too many inputs."
-
-            if "past_key_value_states" in inputs:
-                warnings.warn(
-                    "The `past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = inputs.pop("past_key_value_states")
         else:
             input_ids = inputs
 
-            if "past_key_value_states" in kwargs:
-                warnings.warn(
-                    "The `past_key_value_states` argument is deprecated and will be removed in a future version, use `past_key_values` instead.",
-                    FutureWarning,
-                )
-                past_key_values = kwargs.pop("past_key_value_states")
-
         use_cache = use_cache if use_cache is not None else self.config.use_cache
         output_attentions = output_attentions if output_attentions else self.config.output_attentions
         output_hidden_states = output_hidden_states if output_hidden_states else self.config.output_hidden_states

src/transformers/models/transfo_xl/configuration_transfo_xl.py

@@ -15,9 +15,6 @@
 # limitations under the License.
 """ Transformer XL configuration """
 
-
-import warnings
-
 from ...configuration_utils import PretrainedConfig
 from ...utils import logging
 
@@ -139,13 +136,6 @@ class TransfoXLConfig(PretrainedConfig):
         eos_token_id=0,
         **kwargs
     ):
-        if "tie_weight" in kwargs:
-            warnings.warn(
-                "The config parameter `tie_weight` is deprecated. Please use `tie_word_embeddings` instead.",
-                FutureWarning,
-            )
-            kwargs["tie_word_embeddings"] = kwargs["tie_weight"]
-
         super().__init__(eos_token_id=eos_token_id, **kwargs)
         self.vocab_size = vocab_size
         self.cutoffs = []

src/transformers/models/transfo_xl/modeling_tf_transfo_xl.py

@@ -16,7 +16,6 @@
 """
  TF 2.0 Transformer XL model.
 """
-import warnings
 from dataclasses import dataclass
 from typing import List, Optional, Tuple
 
@@ -865,13 +864,6 @@ class TFTransfoXLLMHeadModel(TFTransfoXLPreTrainedModel):
             return self.crit.out_layers[-1]
         return None
 
-    def reset_length(self, tgt_len, ext_len, mem_len):
-        warnings.warn(
-            "The method `reset_length` is deprecated and will be removed in a future version, use `reset_memory_length` instead.",
-            FutureWarning,
-        )
-        self.transformer.reset_memory_length(mem_len)
-
     def reset_memory_length(self, mem_len):
         self.transformer.reset_memory_length(mem_len)
 

src/transformers/models/transfo_xl/modeling_transfo_xl.py

@@ -17,7 +17,6 @@
  PyTorch Transformer XL model. Adapted from https://github.com/kimiyoung/transformer-xl. In particular
  https://github.com/kimiyoung/transformer-xl/blob/master/pytorch/mem_transformer.py
 """
-import warnings
 from dataclasses import dataclass
 from typing import List, Optional, Tuple
 
@@ -1010,13 +1009,6 @@ class TransfoXLLMHeadModel(TransfoXLPreTrainedModel):
                     else:
                         self.crit.out_projs[i] = self.transformer.word_emb.emb_projs[i]
 
-    def reset_length(self, tgt_len, ext_len, mem_len):
-        warnings.warn(
-            "The method `reset_length` is deprecated and will be removed in a future version, use `reset_memory_length` instead.",
-            FutureWarning,
-        )
-        self.transformer.reset_memory_length(mem_len)
-
     def reset_memory_length(self, mem_len):
         self.transformer.reset_memory_length(mem_len)
 

src/transformers/models/xlm/modeling_tf_xlm.py

@@ -16,9 +16,7 @@
  TF 2.0 XLM model.
 """
 
-
 import itertools
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -997,10 +995,9 @@ class TFXLMForMultipleChoice(TFXLMPreTrainedModel, TFMultipleChoiceLoss):
         )
 
         if lengths is not None:
-            warnings.warn(
+            logger.warn(
                 "The `lengths` parameter cannot be used with the XLM multiple choice models. Please use the "
                 "attention mask instead.",
-                FutureWarning,
             )
             lengths = None
 

src/transformers/models/xlm/modeling_xlm.py

@@ -16,10 +16,8 @@
  PyTorch XLM model.
 """
 
-
 import itertools
 import math
-import warnings
 from dataclasses import dataclass
 from typing import Optional, Tuple
 
@@ -1228,10 +1226,9 @@ class XLMForMultipleChoice(XLMPreTrainedModel):
         )
 
         if lengths is not None:
-            warnings.warn(
+            logger.warn(
                 "The `lengths` parameter cannot be used with the XLM multiple choice models. Please use the "
-                "attention mask instead.",
+                "attention mask instead."
-                FutureWarning,
             )
             lengths = None
 

src/transformers/pipelines.py

@@ -1182,7 +1182,6 @@ class FillMaskPipeline(Pipeline):
         device: int = -1,
         top_k=5,
         task: str = "",
-        **kwargs
     ):
         super().__init__(
             model=model,
@@ -1196,15 +1195,7 @@ class FillMaskPipeline(Pipeline):
         )
 
         self.check_model_type(TF_MODEL_WITH_LM_HEAD_MAPPING if self.framework == "tf" else MODEL_FOR_MASKED_LM_MAPPING)
-
+        self.top_k = top_k
-        if "topk" in kwargs:
-            warnings.warn(
-                "The `topk` argument is deprecated and will be removed in a future version, use `top_k` instead.",
-                FutureWarning,
-            )
-            self.top_k = kwargs.pop("topk")
-        else:
-            self.top_k = top_k
 
     def ensure_exactly_one_mask_token(self, masked_index: np.ndarray):
         numel = np.prod(masked_index.shape)

src/transformers/tokenization_utils.py

@@ -19,7 +19,6 @@
 import itertools
 import re
 import unicodedata
-import warnings
 from typing import Any, Dict, List, Optional, Tuple, Union, overload
 
 from .file_utils import add_end_docstrings
@@ -246,12 +245,6 @@ class PreTrainedTokenizer(PreTrainedTokenizerBase):
         Returns:
             :obj:`List[str]`: The list of tokens.
         """
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            kwargs["is_split_into_words"] = kwargs.pop("is_pretokenized")
         # Simple mapping string => AddedToken for special tokens with specific tokenization behaviors
         all_special_tokens_extended = dict(
             (str(t), t) for t in self.all_special_tokens_extended if isinstance(t, AddedToken)
@@ -448,13 +441,6 @@ class PreTrainedTokenizer(PreTrainedTokenizerBase):
                 "https://github.com/huggingface/transformers/pull/2674"
             )
 
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         first_ids = get_input_ids(text)
         second_ids = get_input_ids(text_pair) if text_pair is not None else None
 
@@ -530,13 +516,6 @@ class PreTrainedTokenizer(PreTrainedTokenizerBase):
                 "transformers.PreTrainedTokenizerFast."
             )
 
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         input_ids = []
         for ids_or_pair_ids in batch_text_or_text_pairs:
             if not isinstance(ids_or_pair_ids, (list, tuple)):

src/transformers/tokenization_utils_base.py

@@ -1532,18 +1532,6 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
 
         super().__init__(**kwargs)
 
-    @property
-    def max_len(self) -> int:
-        """
-        :obj:`int`: **Deprecated** Kept here for backward compatibility. Now renamed to :obj:`model_max_length` to
-        avoid ambiguity.
-        """
-        warnings.warn(
-            "The `max_len` attribute has been deprecated and will be removed in a future version, use `model_max_length` instead.",
-            FutureWarning,
-        )
-        return self.model_max_length
-
     @property
     def max_len_single_sentence(self) -> int:
         """
@@ -2785,15 +2773,6 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
                 and ``convert_tokens_to_ids`` methods.
         """
 
-        if "return_lengths" in kwargs:
-            if verbose:
-                warnings.warn(
-                    "The PreTrainedTokenizerBase.prepare_for_model `return_lengths` parameter is deprecated. "
-                    "Please use `return_length` instead.",
-                    FutureWarning,
-                )
-            return_length = kwargs["return_lengths"]
-
         # Backward compatibility for 'truncation_strategy', 'pad_to_max_length'
         padding_strategy, truncation_strategy, max_length, kwargs = self._get_padding_truncation_strategies(
             padding=padding,

src/transformers/tokenization_utils_fast.py

@@ -19,7 +19,6 @@
 
 import json
 import os
-import warnings
 from collections import defaultdict
 from typing import Any, Dict, List, Optional, Tuple, Union
 
@@ -357,7 +356,6 @@ class PreTrainedTokenizerFast(PreTrainedTokenizerBase):
         return_offsets_mapping: bool = False,
         return_length: bool = False,
         verbose: bool = True,
-        **kwargs
     ) -> BatchEncoding:
 
         if not isinstance(batch_text_or_text_pairs, list):
@@ -365,16 +363,6 @@ class PreTrainedTokenizerFast(PreTrainedTokenizerBase):
                 "batch_text_or_text_pairs has to be a list (got {})".format(type(batch_text_or_text_pairs))
             )
 
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
-        if kwargs:
-            raise ValueError(f"Keyword arguments {kwargs} not recognized.")
-
         # Set the truncation and padding strategy and restore the initial configuration
         self.set_truncation_and_padding(
             padding_strategy=padding_strategy,
@@ -453,12 +441,6 @@ class PreTrainedTokenizerFast(PreTrainedTokenizerBase):
         verbose: bool = True,
         **kwargs
     ) -> BatchEncoding:
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
 
         batched_input = [(text, text_pair)] if text_pair else [text]
         batched_output = self._batch_encode_plus(

src/transformers/trainer.py

@@ -213,8 +213,6 @@ class Trainer:
             containing the optimizer and the scheduler to use. Will default to an instance of
             :class:`~transformers.AdamW` on your model and a scheduler given by
             :func:`~transformers.get_linear_schedule_with_warmup` controlled by :obj:`args`.
-        kwargs:
-            Deprecated keyword arguments.
     """
 
     def __init__(
@@ -229,7 +227,6 @@ class Trainer:
         compute_metrics: Optional[Callable[[EvalPrediction], Dict]] = None,
         callbacks: Optional[List[TrainerCallback]] = None,
         optimizers: Tuple[torch.optim.Optimizer, torch.optim.lr_scheduler.LambdaLR] = (None, None),
-        **kwargs,
     ):
         if args is None:
             logger.info("No `TrainingArguments` passed, using the current path as `output_dir`.")
@@ -262,27 +259,6 @@ class Trainer:
         self.callback_handler = CallbackHandler(callbacks, self.model, self.optimizer, self.lr_scheduler)
         self.add_callback(PrinterCallback if self.args.disable_tqdm else DEFAULT_PROGRESS_CALLBACK)
 
-        # Deprecated arguments
-        if "tb_writer" in kwargs:
-            warnings.warn(
-                "Passing `tb_writer` as a keyword argument is deprecated and won't be possible in a "
-                + "future version. Use `TensorBoardCallback(tb_writer=...)` instead and pass it to the `callbacks`"
-                + "argument",
-                FutureWarning,
-            )
-            tb_writer = kwargs.pop("tb_writer")
-            self.remove_callback(TensorBoardCallback)
-            self.add_callback(TensorBoardCallback(tb_writer=tb_writer))
-        if "prediction_loss_only" in kwargs:
-            warnings.warn(
-                "Passing `prediction_loss_only` as a keyword argument is deprecated and won't be possible in a "
-                + "future version. Use `args.prediction_loss_only` instead. Setting "
-                + f"`args.prediction_loss_only={kwargs['prediction_loss_only']}",
-                FutureWarning,
-            )
-            self.args.prediction_loss_only = kwargs.pop("prediction_loss_only")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
-
         # Will be set to True by `self._setup_loggers()` on first call to `self.log()`.
         self._loggers_initialized = False
 
@@ -294,14 +270,7 @@ class Trainer:
             # We'll find a more elegant and not need to do this in the future.
             self.model.config.xla_device = True
         if not callable(self.data_collator) and callable(getattr(self.data_collator, "collate_batch", None)):
-            self.data_collator = self.data_collator.collate_batch
+            raise ValueError("The `data_collator` should be a simple callable (function, class with `__call__`).")
-            warnings.warn(
-                (
-                    "The `data_collator` should now be a simple callable (function, class with `__call__`), classes "
-                    + "with a `collate_batch` are deprecated and won't be supported in a future version."
-                ),
-                FutureWarning,
-            )
 
         if args.max_steps > 0:
             logger.info("max_steps is given, it will override any value given in num_train_epochs")
@@ -1050,12 +1019,6 @@ class Trainer:
             logs (:obj:`Dict[str, float]`):
                 The values to log.
         """
-        if hasattr(self, "_log"):
-            warnings.warn(
-                "The `_log` method is deprecated and won't be called in a future version, define `log` in your subclass.",
-                FutureWarning,
-            )
-            return self._log(logs)
         if self.state.epoch is not None:
             logs["epoch"] = self.state.epoch
 
@@ -1095,12 +1058,6 @@ class Trainer:
         Return:
             :obj:`torch.Tensor`: The tensor with training loss on this batch.
         """
-        if hasattr(self, "_training_step"):
-            warnings.warn(
-                "The `_training_step` method is deprecated and won't be called in a future version, define `training_step` in your subclass.",
-                FutureWarning,
-            )
-            return self._training_step(model, inputs, self.optimizer)
 
         model.train()
         inputs = self._prepare_inputs(inputs)
@@ -1140,18 +1097,6 @@ class Trainer:
         # We don't use .loss here since the model may return tuples instead of ModelOutput.
         return outputs[0]
 
-    def is_local_master(self) -> bool:
-        """
-        Whether or not this process is the local (e.g., on one machine if training in a distributed fashion on several
-        machines) main process.
-
-        .. warning::
-
-            This method is deprecated, use :meth:`~transformers.Trainer.is_local_process_zero` instead.
-        """
-        warnings.warn("This method is deprecated, use `Trainer.is_local_process_zero()` instead.", FutureWarning)
-        return self.is_local_process_zero()
-
     def is_local_process_zero(self) -> bool:
         """
         Whether or not this process is the local (e.g., on one machine if training in a distributed fashion on several
@@ -1162,18 +1107,6 @@ class Trainer:
         else:
             return self.args.local_rank in [-1, 0]
 
-    def is_world_master(self) -> bool:
-        """
-        Whether or not this process is the global main process (when training in a distributed fashion on several
-        machines, this is only going to be :obj:`True` for one process).
-
-        .. warning::
-
-            This method is deprecated, use :meth:`~transformers.Trainer.is_world_process_zero` instead.
-        """
-        warnings.warn("This method is deprecated, use `Trainer.is_world_process_zero()` instead.", FutureWarning)
-        return self.is_world_process_zero()
-
     def is_world_process_zero(self) -> bool:
         """
         Whether or not this process is the global main process (when training in a distributed fashion on several
@@ -1362,13 +1295,6 @@ class Trainer:
 
         Works both with or without labels.
         """
-        if hasattr(self, "_prediction_loop"):
-            warnings.warn(
-                "The `_prediction_loop` method is deprecated and won't be called in a future version, define `prediction_loop` in your subclass.",
-                FutureWarning,
-            )
-            return self._prediction_loop(dataloader, description, prediction_loss_only=prediction_loss_only)
-
         if not isinstance(dataloader.dataset, collections.abc.Sized):
             raise ValueError("dataset must implement __len__")
         prediction_loss_only = (

src/transformers/trainer_tf.py

@@ -3,7 +3,6 @@
 import datetime
 import math
 import os
-import warnings
 from typing import Callable, Dict, Optional, Tuple
 
 
@@ -66,8 +65,6 @@ class TFTrainer:
             :class:`~transformers.AdamWeightDecay`. The scheduler will default to an instance of
             :class:`tf.keras.optimizers.schedules.PolynomialDecay` if :obj:`args.num_warmup_steps` is 0 else an
             instance of :class:`~transformers.WarmUp`.
-        kwargs:
-            Deprecated keyword arguments.
     """
 
     def __init__(
@@ -82,7 +79,6 @@ class TFTrainer:
             None,
             None,
         ),
-        **kwargs,
     ):
         assert parse(tf.__version__).release >= (2, 2, 0), (
             "You need to run the TensorFlow trainer with at least the version 2.2.0, your version is %r "
@@ -98,13 +94,6 @@ class TFTrainer:
         self.gradient_accumulator = GradientAccumulator()
         self.global_step = 0
         self.epoch_logging = 0
-        if "prediction_loss_only" in kwargs:
-            warnings.warn(
-                "Passing `prediction_loss_only` as a keyword argument is deprecated and won't be possible in a future version. Use `args.prediction_loss_only` instead.",
-                FutureWarning,
-            )
-            self.args.prediction_loss_only = kwargs.pop("prediction_loss_only")
-        assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
 
         if tb_writer is not None:
             self.tb_writer = tb_writer
@@ -249,12 +238,6 @@ class TFTrainer:
             WANDB_DISABLED:
                 (Optional): boolean - defaults to false, set to "true" to disable wandb entirely.
         """
-        if hasattr(self, "_setup_wandb"):
-            warnings.warn(
-                "The `_setup_wandb` method is deprecated and won't be called in a future version, define `setup_wandb` in your subclass.",
-                FutureWarning,
-            )
-            return self._setup_wandb()
 
         logger.info('Automatic Weights & Biases logging enabled, to disable set os.environ["WANDB_DISABLED"] = "true"')
         combined_dict = {**self.model.config.to_dict(), **self.args.to_sanitized_dict()}
@@ -304,14 +287,6 @@ class TFTrainer:
 
         Works both with or without labels.
         """
-        if hasattr(self, "_prediction_loop"):
-            warnings.warn(
-                "The `_prediction_loop` method is deprecated and won't be called in a future version, define `prediction_loop` in your subclass.",
-                FutureWarning,
-            )
-            return self._prediction_loop(
-                dataset, steps, num_examples, description, prediction_loss_only=prediction_loss_only
-            )
 
         prediction_loss_only = (
             prediction_loss_only if prediction_loss_only is not None else self.args.prediction_loss_only
@@ -393,12 +368,6 @@ class TFTrainer:
             logs (:obj:`Dict[str, float]`):
                 The values to log.
         """
-        if hasattr(self, "_log"):
-            warnings.warn(
-                "The `_log` method is deprecated and won't be called in a future version, define `log` in your subclass.",
-                FutureWarning,
-            )
-            return self._log(logs)
         logs["epoch"] = self.epoch_logging
 
         if self.tb_writer:
@@ -733,12 +702,6 @@ class TFTrainer:
         Returns:
             A tuple of two :obj:`tf.Tensor`: The loss and logits.
         """
-        if hasattr(self, "_run_model"):
-            warnings.warn(
-                "The `_run_model` method is deprecated and won't be called in a future version, define `run_model` in your subclass.",
-                FutureWarning,
-            )
-            return self._run_model(features, labels, training)
 
         if self.args.past_index >= 0 and getattr(self, "_past", None) is not None:
             features["mems"] = self._past

src/transformers/training_args.py

@@ -1,7 +1,6 @@
 import dataclasses
 import json
 import os
-import warnings
 from dataclasses import dataclass, field
 from enum import Enum
 from typing import Any, Dict, List, Optional, Tuple
@@ -198,10 +197,6 @@ class TrainingArguments:
     do_train: bool = field(default=False, metadata={"help": "Whether to run training."})
     do_eval: bool = field(default=None, metadata={"help": "Whether to run eval on the dev set."})
     do_predict: bool = field(default=False, metadata={"help": "Whether to run predictions on the test set."})
-    evaluate_during_training: bool = field(
-        default=False,
-        metadata={"help": "Run evaluation during training at each logging step."},
-    )
     evaluation_strategy: EvaluationStrategy = field(
         default="no",
         metadata={"help": "Run evaluation during training at each logging step."},
@@ -340,12 +335,6 @@ class TrainingArguments:
     def __post_init__(self):
         if self.disable_tqdm is None:
             self.disable_tqdm = logger.getEffectiveLevel() > logging.WARN
-        if self.evaluate_during_training is True:
-            self.evaluation_strategy = EvaluationStrategy.STEPS
-            warnings.warn(
-                "The `evaluate_during_training` argument is deprecated in favor of `evaluation_strategy` (which has more options)",
-                FutureWarning,
-            )
         self.evaluation_strategy = EvaluationStrategy(self.evaluation_strategy)
         if self.do_eval is False and self.evaluation_strategy != EvaluationStrategy.NO:
             self.do_eval = True

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/tokenization_{{cookiecutter.lowercase_modelname}}.py

@@ -73,7 +73,6 @@ class {{cookiecutter.camelcase_modelname}}TokenizerFast(BertTokenizerFast):
     max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
     pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
 {%- elif cookiecutter.tokenizer_type == "Standalone" %}
-import warnings
 from typing import List, Optional
 
 from tokenizers import ByteLevelBPETokenizer
@@ -234,13 +233,6 @@ class {{cookiecutter.camelcase_modelname}}Tokenizer(PreTrainedTokenizer):
         return len(cls + token_ids_0 + sep + sep + token_ids_1 + sep) * [0]
 
     def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
         add_prefix_space = kwargs.pop("add_prefix_space", self.add_prefix_space)
         if (is_split_into_words or add_prefix_space) and (len(text) > 0 and not text[0].isspace()):
             text = " " + text
@@ -285,29 +277,6 @@ class {{cookiecutter.camelcase_modelname}}TokenizerFast(PreTrainedTokenizerFast)
         )
         self.add_prefix_space = add_prefix_space
 
-    def _batch_encode_plus(self, *args, **kwargs) -> BatchEncoding:
-        is_split_into_words = None
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-            is_split_into_words = kwargs.pop("is_pretokenized")
-
-        is_split_into_words = kwargs.get("is_split_into_words", False) if is_split_into_words is None else is_split_into_words
-        return super()._batch_encode_plus(*args, **kwargs)
-
-    def _encode_plus(self, *args, **kwargs) -> BatchEncoding:
-        is_split_into_words = None
-        if "is_pretokenized" in kwargs:
-            warnings.warn(
-                "`is_pretokenized` is deprecated and will be removed in a future version, use `is_split_into_words` instead.",
-                FutureWarning,
-            )
-        is_split_into_words = kwargs.get("is_split_into_words", False) if is_split_into_words is None else is_split_into_words
-        return super()._encode_plus(*args, **kwargs)
-
-
     def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
         output = [self.bos_token_id] + token_ids_0 + [self.eos_token_id]
         if token_ids_1 is None:

tests/test_modeling_gpt2.py

@@ -213,7 +213,9 @@ class GPT2ModelTester:
         next_token_type_ids = torch.cat([token_type_ids, next_token_types], dim=-1)
 
         output_from_no_past = model(next_input_ids, token_type_ids=next_token_type_ids)["last_hidden_state"]
-        output_from_past = model(next_tokens, token_type_ids=next_token_types, past=past)["last_hidden_state"]
+        output_from_past = model(next_tokens, token_type_ids=next_token_types, past_key_values=past)[
+            "last_hidden_state"
+        ]
 
         # select random slice
         random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
@@ -255,7 +257,7 @@ class GPT2ModelTester:
 
         # get two different outputs
         output_from_no_past = model(next_input_ids, attention_mask=attn_mask)["last_hidden_state"]
-        output_from_past = model(next_tokens, past=past, attention_mask=attn_mask)["last_hidden_state"]
+        output_from_past = model(next_tokens, past_key_values=past, attention_mask=attn_mask)["last_hidden_state"]
 
         # select random slice
         random_slice_idx = ids_tensor((1,), output_from_past.shape[-1]).item()
@@ -286,7 +288,9 @@ class GPT2ModelTester:
         next_token_type_ids = torch.cat([token_type_ids, next_token_types], dim=-1)
 
         output_from_no_past = model(next_input_ids, token_type_ids=next_token_type_ids)["last_hidden_state"]
-        output_from_past = model(next_tokens, token_type_ids=next_token_types, past=past)["last_hidden_state"]
+        output_from_past = model(next_tokens, token_type_ids=next_token_types, past_key_values=past)[
+            "last_hidden_state"
+        ]
         self.parent.assertTrue(output_from_past.shape[1] == next_tokens.shape[1])
 
         # select random slice

tests/test_pipelines_fill_mask.py

@@ -1,7 +1,5 @@
 import unittest
 
-import pytest
-
 from transformers import pipeline
 from transformers.testing_utils import require_tf, require_torch, slow
 
@@ -53,13 +51,6 @@ class FillMaskPipelineTests(MonoInputPipelineCommonMixin, unittest.TestCase):
     ]
     expected_check_keys = ["sequence"]
 
-    @require_torch
-    def test_torch_topk_deprecation(self):
-        # At pipeline initialization only it was not enabled at pipeline
-        # call site before
-        with pytest.warns(FutureWarning, match=r".*use `top_k`.*"):
-            pipeline(task="fill-mask", model=self.small_models[0], topk=1)
-
     @require_torch
     def test_torch_fill_mask(self):
         valid_inputs = "My name is <mask>"

tests/test_tokenization_auto.py

@@ -83,7 +83,7 @@ class AutoTokenizerTest(unittest.TestCase):
             else:
                 self.assertEqual(tokenizer.do_lower_case, False)
 
-            self.assertEqual(tokenizer.max_len, 512)
+            self.assertEqual(tokenizer.model_max_length, 512)
 
     @require_tokenizers
     def test_tokenizer_identifier_non_existent(self):