Doc fixes in preparation for the docstyle PR (#8061)

* Fixes in preparation for doc styling

* More fixes

* Better syntax

* Fixes

* Style

* More fixes

* More fixes

src/transformers/modeling_tf_openai.py

@@ -625,7 +625,7 @@ class TFOpenAIGPTDoubleHeadsModel(TFOpenAIGPTPreTrainedModel):
         training=False,
     ):
         r"""
-        mc_token_ids (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, num_choices)`, `optional`, default to index of the last token of the input)
+        mc_token_ids (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, num_choices)`, `optional`, default to index of the last token of the input):
             Index of the classification token in each input sequence.
             Selected in the range ``[0, input_ids.size(-1) - 1]``.
 

src/transformers/modeling_tf_roberta.py

@@ -111,8 +111,11 @@ class TFRobertaEmbeddings(tf.keras.layers.Layer):
         """Replace non-padding symbols with their position numbers. Position numbers begin at
         padding_idx+1. Padding symbols are ignored. This is modified from fairseq's
         `utils.make_positions`.
-        :param tf.Tensor x:
-        :return tf.Tensor:
+
+        Args:
+            x: tf.Tensor
+
+        Returns: tf.Tensor
         """
         mask = tf.cast(tf.math.not_equal(x, self.padding_idx), dtype=tf.int32)
         incremental_indicies = tf.math.cumsum(mask, axis=1) * mask
@@ -122,8 +125,11 @@ class TFRobertaEmbeddings(tf.keras.layers.Layer):
     def create_position_ids_from_inputs_embeds(self, inputs_embeds):
         """We are provided embeddings directly. We cannot infer which are padded so just generate
         sequential position ids.
-        :param tf.Tensor inputs_embeds:
-        :return tf.Tensor:
+
+        Args:
+            inputs_embeds: tf.Tensor
+
+        Returns: tf.Tensor
         """
         seq_length = shape_list(inputs_embeds)[1]
         position_ids = tf.range(self.padding_idx + 1, seq_length + self.padding_idx + 1, dtype=tf.int32)[tf.newaxis, :]

src/transformers/modeling_tf_xlnet.py

@@ -1718,120 +1718,3 @@ class TFXLNetForQuestionAnsweringSimple(TFXLNetPreTrainedModel, TFQuestionAnswer
             hidden_states=transformer_outputs.hidden_states,
             attentions=transformer_outputs.attentions,
         )
-
-
-# @add_start_docstrings("""XLNet Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of
-#     the hidden-states output to compute `span start logits` and `span end logits`). """,
-#     XLNET_START_DOCSTRING, XLNET_INPUTS_DOCSTRING)
-# class TFXLNetForQuestionAnswering(TFXLNetPreTrainedModel):
-#     r"""
-#     Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
-#         **start_top_log_probs**: (`optional`, returned if ``start_positions`` or ``end_positions`` is not provided)
-#             ``tf.Tensor`` of shape ``(batch_size, config.start_n_top)``
-#             Log probabilities for the top config.start_n_top start token possibilities (beam-search).
-#         **start_top_index**: (`optional`, returned if ``start_positions`` or ``end_positions`` is not provided)
-#             ``tf.Tensor`` of shape ``(batch_size, config.start_n_top)``
-#             Indices for the top config.start_n_top start token possibilities (beam-search).
-#         **end_top_log_probs**: (`optional`, returned if ``start_positions`` or ``end_positions`` is not provided)
-#             ``tf.Tensor`` of shape ``(batch_size, config.start_n_top * config.end_n_top)``
-#             Log probabilities for the top ``config.start_n_top * config.end_n_top`` end token possibilities (beam-search).
-#         **end_top_index**: (`optional`, returned if ``start_positions`` or ``end_positions`` is not provided)
-#             ``tf.Tensor`` of shape ``(batch_size, config.start_n_top * config.end_n_top)``
-#             Indices for the top ``config.start_n_top * config.end_n_top`` end token possibilities (beam-search).
-#         **cls_logits**: (`optional`, returned if ``start_positions`` or ``end_positions`` is not provided)
-#             ``tf.Tensor`` of shape ``(batch_size,)``
-#             Log probabilities for the ``is_impossible`` label of the answers.
-#         **mems**:
-#             list of ``tf.Tensor`` (one for each layer):
-#             that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model
-#             if config.mem_len > 0 else tuple of None. Can be used to speed up sequential decoding and attend to longer context.
-#             See details in the docstring of the `mems` input above.
-#         **hidden_states**: (`optional`, returned when ``output_hidden_states=True`` is passed or when ``config.output_hidden_states=True``)
-#             list of ``tf.Tensor`` (one for the output of each layer + the output of the embeddings)
-#             of shape ``(batch_size, sequence_length, hidden_size)``:
-#             Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-#         **attentions**: (`optional`, returned when ``output_attentions=True``)
-#             list of ``tf.Tensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
-#             Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
-
-#     Examples::
-
-#         # For example purposes. Not runnable.
-#         tokenizer = XLMTokenizer.from_pretrained('xlm-mlm-en-2048')
-#         model = XLMForQuestionAnswering.from_pretrained('xlnet-large-cased')
-#         input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
-#         start_positions = tf.constant([1])
-#         end_positions = tf.constant([3])
-#         outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)
-#         loss, start_scores, end_scores = outputs[:2]
-
-#     """
-#     def __init__(self, config, *inputs, **kwargs):
-#         super().__init__(config, *inputs, **kwargs)
-#         self.start_n_top = config.start_n_top
-#         self.end_n_top = config.end_n_top
-
-#         self.transformer = TFXLNetMainLayer(config, name='transformer')
-#         self.start_logits = TFPoolerStartLogits(config, name='start_logits')
-#         self.end_logits = TFPoolerEndLogits(config, name='end_logits')
-#         self.answer_class = TFPoolerAnswerClass(config, name='answer_class')
-
-#     def call(self, inputs, training=False):
-#         transformer_outputs = self.transformer(inputs, training=training)
-#         hidden_states = transformer_outputs[0]
-#         start_logits = self.start_logits(hidden_states, p_mask=p_mask)
-
-#         outputs = transformer_outputs[1:]  # Keep mems, hidden states, attentions if there are in it
-
-#         if start_positions is not None and end_positions is not None:
-#             # If we are on multi-GPU, let's remove the dimension added by batch splitting
-#             for x in (start_positions, end_positions, cls_index, is_impossible):
-#                 if x is not None and x.dim() > 1:
-#                     x.squeeze_(-1)
-
-#             # during training, compute the end logits based on the ground truth of the start position
-#             end_logits = self.end_logits(hidden_states, start_positions=start_positions, p_mask=p_mask)
-
-#             loss_fct = CrossEntropyLoss()
-#             start_loss = loss_fct(start_logits, start_positions)
-#             end_loss = loss_fct(end_logits, end_positions)
-#             total_loss = (start_loss + end_loss) / 2
-
-#             if cls_index is not None and is_impossible is not None:
-#                 # Predict answerability from the representation of CLS and START
-#                 cls_logits = self.answer_class(hidden_states, start_positions=start_positions, cls_index=cls_index)
-#                 loss_fct_cls = nn.BCEWithLogitsLoss()
-#                 cls_loss = loss_fct_cls(cls_logits, is_impossible)
-
-#                 # note(zhiliny): by default multiply the loss by 0.5 so that the scale is comparable to start_loss and end_loss
-#                 total_loss += cls_loss * 0.5
-
-#             outputs = (total_loss,) + outputs
-
-#         else:
-#             # during inference, compute the end logits based on beam search
-#             bsz, slen, hsz = hidden_states.size()
-#             start_log_probs = F.softmax(start_logits, dim=-1) # shape (bsz, slen)
-
-#             start_top_log_probs, start_top_index = torch.topk(start_log_probs, self.start_n_top, dim=-1) # shape (bsz, start_n_top)
-#             start_top_index_exp = start_top_index.unsqueeze(-1).expand(-1, -1, hsz) # shape (bsz, start_n_top, hsz)
-#             start_states = torch.gather(hidden_states, -2, start_top_index_exp) # shape (bsz, start_n_top, hsz)
-#             start_states = start_states.unsqueeze(1).expand(-1, slen, -1, -1) # shape (bsz, slen, start_n_top, hsz)
-
-#             hidden_states_expanded = hidden_states.unsqueeze(2).expand_as(start_states) # shape (bsz, slen, start_n_top, hsz)
-#             p_mask = p_mask.unsqueeze(-1) if p_mask is not None else None
-#             end_logits = self.end_logits(hidden_states_expanded, start_states=start_states, p_mask=p_mask)
-#             end_log_probs = F.softmax(end_logits, dim=1) # shape (bsz, slen, start_n_top)
-
-#             end_top_log_probs, end_top_index = torch.topk(end_log_probs, self.end_n_top, dim=1) # shape (bsz, end_n_top, start_n_top)
-#             end_top_log_probs = end_top_log_probs.view(-1, self.start_n_top * self.end_n_top)
-#             end_top_index = end_top_index.view(-1, self.start_n_top * self.end_n_top)
-
-#             start_states = torch.einsum("blh,bl->bh", hidden_states, start_log_probs)  # get the representation of START as weighted sum of hidden states
-#             cls_logits = self.answer_class(hidden_states, start_states=start_states, cls_index=cls_index)  # Shape (batch size,): one single `cls_logits` for each sample
-
-#             outputs = (start_top_log_probs, start_top_index, end_top_log_probs, end_top_index, cls_logits) + outputs
-
-#         # return start_top_log_probs, start_top_index, end_top_log_probs, end_top_index, cls_logits
-#         # or (if labels are provided) (total_loss,)
-#         return outputs

src/transformers/modeling_xlnet.py

@@ -1487,7 +1487,7 @@ class XLNetForSequenceClassification(XLNetPreTrainedModel):
         return_dict=None,
     ):
         r"""
-        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`)
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
             Labels for computing the sequence classification/regression loss.
             Indices should be in ``[0, ..., config.num_labels - 1]``.
             If ``config.num_labels == 1`` a regression loss is computed (Mean-Square loss),

src/transformers/testing_utils.py

@@ -352,7 +352,7 @@ class CaptureStd:
     - out - capture stdout: True/False, default True
     - err - capture stdout: True/False, default True
 
-    Examples:
+    Examples::
 
         with CaptureStdout() as cs:
             print("Secret message")
@@ -444,7 +444,7 @@ class CaptureLogger:
     Results:
         The captured output is available via `self.out`
 
-    Example:
+    Example::
 
         >>> from transformers import logging
         >>> from transformers.testing_utils import CaptureLogger
@@ -485,22 +485,34 @@ class TestCasePlus(unittest.TestCase):
     of test, unless `after=False`.
 
     # 1. create a unique temp dir, `tmp_dir` will contain the path to the created temp dir
+
+    ::
+
         def test_whatever(self):
             tmp_dir = self.get_auto_remove_tmp_dir()
 
     # 2. create a temp dir of my choice and delete it at the end - useful for debug when you want to
     # monitor a specific directory
+
+    ::
+
         def test_whatever(self):
             tmp_dir = self.get_auto_remove_tmp_dir(tmp_dir="./tmp/run/test")
 
     # 3. create a temp dir of my choice and do not delete it at the end - useful for when you want
     # to look at the temp results
+
+    ::
+
         def test_whatever(self):
             tmp_dir = self.get_auto_remove_tmp_dir(tmp_dir="./tmp/run/test", after=False)
 
     # 4. create a temp dir of my choice and ensure to delete it right away - useful for when you
     # disabled deletion in the previous test run and want to make sure the that tmp dir is empty
     # before the new test is run
+
+    ::
+
         def test_whatever(self):
             tmp_dir = self.get_auto_remove_tmp_dir(tmp_dir="./tmp/run/test", before=True)
 

src/transformers/tokenization_bertweet.py

@@ -488,6 +488,7 @@ domains and tasks. The basic logic is this:
 
 # This particular element is used in a couple ways, so we define it
 # with a name:
+# docstyle-ignore
 EMOTICONS = r"""
     (?:
       [<>]?
@@ -505,7 +506,7 @@ EMOTICONS = r"""
 
 # URL pattern due to John Gruber, modified by Tom Winzig. See
 # https://gist.github.com/winzig/8894715
-
+# docstyle-ignore
 URLS = r"""			# Capture 1: entire matched URL
   (?:
   https?:				# URL protocol and colon
@@ -549,6 +550,7 @@ URLS = r"""			# Capture 1: entire matched URL
   )
 """
 
+# docstyle-ignore
 # The components of the tokenizer:
 REGEXPS = (
     URLS,
@@ -628,18 +630,16 @@ def _replace_html_entities(text, keep=(), remove_illegal=True, encoding="utf-8")
     Remove entities from text by converting them to their
     corresponding unicode character.
 
-    :param text: a unicode string or a byte string encoded in the given
-    `encoding` (which defaults to 'utf-8').
-
-    :param list keep:  list of entity names which should not be replaced.\
-    This supports both numeric entities (``&#nnnn;`` and ``&#hhhh;``)
+    Args:
+        text:
+            A unicode string or a byte string encoded in the given `encoding` (which defaults to 'utf-8').
+        keep (list):
+            List of entity names which should not be replaced. This supports both numeric entities (``&#nnnn;`` and ``&#hhhh;``)
             and named entities (such as ``&nbsp;`` or ``&gt;``).
+        remove_illegal (bool):
+            If `True`, entities that can't be converted are removed. Otherwise, entities that can't be converted are kept "as is".
 
-    :param bool remove_illegal: If `True`, entities that can't be converted are\
-    removed. Otherwise, entities that can't be converted are kept "as
-    is".
-
-    :returns: A unicode string with the entities removed.
+    Returns: A unicode string with the entities removed.
 
     See https://github.com/scrapy/w3lib/blob/master/w3lib/html.py
 
@@ -688,16 +688,16 @@ def _replace_html_entities(text, keep=(), remove_illegal=True, encoding="utf-8")
 
 class TweetTokenizer:
     r"""
-    Tokenizer for tweets.
+    Examples::
 
+        >>> # Tokenizer for tweets.
         >>> from nltk.tokenize import TweetTokenizer
         >>> tknzr = TweetTokenizer()
         >>> s0 = "This is a cooool #dummysmiley: :-) :-P <3 and some arrows < > -> <--"
         >>> tknzr.tokenize(s0)
         ['This', 'is', 'a', 'cooool', '#dummysmiley', ':', ':-)', ':-P', '<3', 'and', 'some', 'arrows', '<', '>', '->', '<--']
 
-    Examples using `strip_handles` and `reduce_len parameters`:
-
+        >>> # Examples using `strip_handles` and `reduce_len parameters`:
         >>> tknzr = TweetTokenizer(strip_handles=True, reduce_len=True)
         >>> s1 = '@remy: This is waaaaayyyy too much for you!!!!!!'
         >>> tknzr.tokenize(s1)
@@ -711,10 +711,11 @@ class TweetTokenizer:
 
     def tokenize(self, text):
         """
-        :param text: str
-        :rtype: list(str)
-        :return: a tokenized list of strings; concatenating this list returns\
-        the original string if `preserve_case=False`
+        Args:
+            text: str
+
+        Returns: list(str)
+            A tokenized list of strings; concatenating this list returns the original string if `preserve_case=False`
         """
         # Fix HTML character entities:
         text = _replace_html_entities(text)

src/transformers/tokenization_deberta.py

@@ -628,13 +628,16 @@ class DebertaTokenizer(PreTrainedTokenizer):
 
     def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         """
-        Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
-        A BERT sequence pair mask has the following format:
-        0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
-        | first sequence    | second sequence
+        Create a mask from the two sequences passed to be used in a sequence-pair classification task.
+        A DeBERTa sequence pair mask has the following format:
+
+        ::
+
+            0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+            | first sequence    | second sequence |
+
+        If :obj:`token_ids_1` is :obj:`None`, this method only returns the first portion of the mask (0s).
 
-        if token_ids_1 is None, only returns the first portion of the mask (0's).
-        ~
         Args:
             token_ids_0 (:obj:`List[int]`):
                 List of IDs.