Fix template for inputs docstrings (#12976)

src/transformers/models/big_bird/modeling_big_bird.py

@@ -1789,7 +1789,7 @@ BIG_BIRD_START_DOCSTRING = r"""
 
 BIG_BIRD_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.BigBirdTokenizer`. See
@@ -1797,14 +1797,14 @@ BIG_BIRD_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -1812,7 +1812,7 @@ BIG_BIRD_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
@@ -1823,7 +1823,7 @@ BIG_BIRD_INPUTS_DOCSTRING = r"""
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -1967,7 +1967,7 @@ class BigBirdModel(BigBirdPreTrainedModel):
         self.attention_type = value
         self.encoder.set_attention_type(value)
 
-    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -2374,7 +2374,7 @@ class BigBirdForMaskedLM(BigBirdPreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.cls.predictions.decoder = new_embeddings
 
-    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -2832,7 +2832,7 @@ class BigBirdForTokenClassification(BigBirdPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -2940,7 +2940,7 @@ class BigBirdForQuestionAnswering(BigBirdPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(BIG_BIRD_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint="google/bigbird-base-trivia-itc",

src/transformers/models/canine/modeling_canine.py

@@ -928,7 +928,7 @@ CANINE_START_DOCSTRING = r"""
 
 CANINE_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.CanineTokenizer`. See
@@ -936,14 +936,14 @@ CANINE_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -951,7 +951,7 @@ CANINE_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
@@ -962,7 +962,7 @@ CANINE_INPUTS_DOCSTRING = r"""
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -1088,7 +1088,7 @@ class CanineModel(CaninePreTrainedModel):
         # `repeated`: [batch_size, char_seq_len, molecule_hidden_size]
         return torch.cat([repeated, remainder_repeated], dim=-2)
 
-    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1458,7 +1458,7 @@ class CanineForTokenClassification(CaninePreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1545,7 +1545,7 @@ class CanineForQuestionAnswering(CaninePreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(CANINE_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/convbert/modeling_convbert.py

@@ -689,7 +689,7 @@ CONVBERT_START_DOCSTRING = r"""
 
 CONVBERT_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.ConvBertTokenizer`. See
@@ -697,7 +697,7 @@ CONVBERT_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
 
@@ -705,7 +705,7 @@ CONVBERT_INPUTS_DOCSTRING = r"""
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -714,7 +714,7 @@ CONVBERT_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
@@ -726,7 +726,7 @@ CONVBERT_INPUTS_DOCSTRING = r"""
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -1163,7 +1163,7 @@ class ConvBertForTokenClassification(ConvBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(CONVBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(CONVBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1250,7 +1250,7 @@ class ConvBertForQuestionAnswering(ConvBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(CONVBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(CONVBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/deberta/modeling_deberta.py

@@ -794,7 +794,7 @@ DEBERTA_START_DOCSTRING = r"""
 
 DEBERTA_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.DebertaTokenizer`. See
@@ -802,14 +802,14 @@ DEBERTA_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -817,12 +817,12 @@ DEBERTA_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
             `What are position IDs? <../glossary.html#position-ids>`_
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.

src/transformers/models/deberta_v2/modeling_deberta_v2.py

@@ -915,7 +915,7 @@ DEBERTA_START_DOCSTRING = r"""
 
 DEBERTA_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.DebertaV2Tokenizer`. See
@@ -923,14 +923,14 @@ DEBERTA_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -938,12 +938,12 @@ DEBERTA_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
             `What are position IDs? <../glossary.html#position-ids>`_
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.

src/transformers/models/deit/modeling_deit.py

@@ -466,7 +466,7 @@ class DeiTModel(DeiTPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,
@@ -570,7 +570,7 @@ class DeiTForImageClassification(DeiTPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=SequenceClassifierOutput, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,
@@ -707,7 +707,7 @@ class DeiTForImageClassificationWithTeacher(DeiTPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_start_docstrings_to_model_forward(DEIT_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=DeiTForImageClassificationWithTeacherOutput, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,

src/transformers/models/ibert/modeling_ibert.py

@@ -774,7 +774,7 @@ class IBertModel(IBertPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(IBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(IBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/mpnet/modeling_mpnet.py

@@ -509,7 +509,7 @@ class MPNetModel(MPNetPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(MPNET_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(MPNET_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -867,7 +867,7 @@ class MPNetForTokenClassification(MPNetPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(MPNET_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(MPNET_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/rembert/modeling_rembert.py

@@ -680,7 +680,7 @@ REMBERT_START_DOCSTRING = r"""
 
 REMBERT_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.RemBertTokenizer`. See
@@ -688,14 +688,14 @@ REMBERT_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -703,7 +703,7 @@ REMBERT_INPUTS_DOCSTRING = r"""
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0,
             config.max_position_embeddings - 1]``.
 
@@ -714,7 +714,7 @@ REMBERT_INPUTS_DOCSTRING = r"""
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -772,7 +772,7 @@ class RemBertModel(RemBertPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint="rembert",
@@ -925,7 +925,7 @@ class RemBertForMaskedLM(RemBertPreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.cls.predictions.decoder = new_embeddings
 
-    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint="rembert",
@@ -1343,7 +1343,7 @@ class RemBertForTokenClassification(RemBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint="rembert",
@@ -1431,7 +1431,7 @@ class RemBertForQuestionAnswering(RemBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(REMBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint="rembert",

src/transformers/models/roberta/modeling_roberta.py

@@ -730,7 +730,7 @@ class RobertaModel(RobertaPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(ROBERTA_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(ROBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/roformer/modeling_roformer.py

@@ -744,7 +744,7 @@ ROFORMER_START_DOCSTRING = r"""
 
 ROFORMER_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.RoFormerTokenizer`. See
@@ -752,14 +752,14 @@ ROFORMER_INPUTS_DOCSTRING = r"""
             details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -773,7 +773,7 @@ ROFORMER_INPUTS_DOCSTRING = r"""
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -832,7 +832,7 @@ class RoFormerModel(RoFormerPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -981,7 +981,7 @@ class RoFormerForMaskedLM(RoFormerPreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.cls.predictions.decoder = new_embeddings
 
-    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1412,7 +1412,7 @@ class RoFormerForTokenClassification(RoFormerPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1499,7 +1499,7 @@ class RoFormerForQuestionAnswering(RoFormerPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(ROFORMER_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/squeezebert/modeling_squeezebert.py

@@ -569,7 +569,7 @@ class SqueezeBertModel(SqueezeBertPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -662,7 +662,7 @@ class SqueezeBertForMaskedLM(SqueezeBertPreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.cls.predictions.decoder = new_embeddings
 
-    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -741,7 +741,7 @@ class SqueezeBertForSequenceClassification(SqueezeBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -839,7 +839,7 @@ class SqueezeBertForMultipleChoice(SqueezeBertPreTrainedModel):
         self.init_weights()
 
     @add_start_docstrings_to_model_forward(
-        SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, num_choices, sequence_length)")
+        SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length")
     )
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
@@ -932,7 +932,7 @@ class SqueezeBertForTokenClassification(SqueezeBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1019,7 +1019,7 @@ class SqueezeBertForQuestionAnswering(SqueezeBertPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(SQUEEZEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,

src/transformers/models/vit/modeling_vit.py

@@ -452,7 +452,7 @@ class ViTModel(ViTPreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,
@@ -555,7 +555,7 @@ class ViTForImageClassification(ViTPreTrainedModel):
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_start_docstrings_to_model_forward(VIT_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=SequenceClassifierOutput, config_class=_CONFIG_FOR_DOC)
     def forward(
         self,

src/transformers/models/xlnet/modeling_xlnet.py

@@ -861,7 +861,7 @@ XLNET_START_DOCSTRING = r"""
 
 XLNET_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.XLNetTokenizer`. See

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

@@ -696,7 +696,7 @@ class {{cookiecutter.camelcase_modelname}}PreTrainedModel(PreTrainedModel):
 
 {{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING = r"""
     Args:
-        input_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`):
+        input_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`):
             Indices of input sequence tokens in the vocabulary.
 
             Indices can be obtained using :class:`transformers.{{cookiecutter.camelcase_modelname}}Tokenizer`.
@@ -704,14 +704,14 @@ class {{cookiecutter.camelcase_modelname}}PreTrainedModel(PreTrainedModel):
             :func:`transformers.PreTrainedTokenizer.__call__` for details.
 
             `What are input IDs? <../glossary.html#input-ids>`__
-        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`{0}`, `optional`):
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`({0})`, `optional`):
             Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``:
 
             - 1 for tokens that are **not masked**,
             - 0 for tokens that are **masked**.
 
             `What are attention masks? <../glossary.html#attention-mask>`__
-        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0,
             1]``:
 
@@ -719,7 +719,7 @@ class {{cookiecutter.camelcase_modelname}}PreTrainedModel(PreTrainedModel):
             - 1 corresponds to a `sentence B` token.
 
             `What are token type IDs? <../glossary.html#token-type-ids>`_
-        position_ids (:obj:`torch.LongTensor` of shape :obj:`{0}`, `optional`):
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`({0})`, `optional`):
             Indices of positions of each input sequence tokens in the position embeddings.
             Selected in the range ``[0, config.max_position_embeddings - 1]``.
 
@@ -730,7 +730,7 @@ class {{cookiecutter.camelcase_modelname}}PreTrainedModel(PreTrainedModel):
             - 1 indicates the head is **not masked**,
             - 0 indicates the head is **masked**.
 
-        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`):
+        inputs_embeds (:obj:`torch.FloatTensor` of shape :obj:`({0}, hidden_size)`, `optional`):
             Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
@@ -788,7 +788,7 @@ class {{cookiecutter.camelcase_modelname}}Model({{cookiecutter.camelcase_modelna
         for layer, heads in heads_to_prune.items():
             self.encoder.layer[layer].attention.prune_heads(heads)
 
-    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -947,7 +947,7 @@ class {{cookiecutter.camelcase_modelname}}ForMaskedLM({{cookiecutter.camelcase_m
     def set_output_embeddings(self, new_embeddings):
         self.cls.predictions.decoder = new_embeddings
 
-    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1385,7 +1385,7 @@ class {{cookiecutter.camelcase_modelname}}ForTokenClassification({{cookiecutter.
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,
@@ -1472,7 +1472,7 @@ class {{cookiecutter.camelcase_modelname}}ForQuestionAnswering({{cookiecutter.ca
 
         self.init_weights()
 
-    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    @add_start_docstrings_to_model_forward({{cookiecutter.uppercase_modelname}}_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         tokenizer_class=_TOKENIZER_FOR_DOC,
         checkpoint=_CHECKPOINT_FOR_DOC,