Upgrade styler to better handle lists (#9423)

* Add missing lines before a new list.

* Update doc styler and restyle some files.

* Fix docstrings of LED and Longformer

docs/source/model_doc/led.rst

@@ -40,7 +40,6 @@ Tips:
   *Longformer*'s *chunked self-attention* layer. :class:`~transformers.LEDTokenizer` is an alias of
   :class:`~transformers.BartTokenizer`.
 - LED works very well on long-range *sequence-to-sequence* tasks where the ``input_ids`` largely exceed a length of
-
   1024 tokens.
 - LED pads the ``input_ids`` to be a multiple of ``config.attention_window`` if required. Therefore a small speed-up is
   gained, when :class:`~transformers.LEDTokenizer` is used with the ``pad_to_multiple_of`` argument.

src/transformers/models/dpr/modeling_tf_dpr.py

@@ -443,16 +443,22 @@ TF_DPR_START_DOCSTRING = r"""
 
     .. note::
 
-        TF 2.0 models accepts two formats as inputs: - having all inputs as keyword arguments (like PyTorch models), or
-        - having all inputs as a list, tuple or dict in the first positional arguments. This second option is useful
-        when using :meth:`tf.keras.Model.fit` method which currently requires having all the tensors in the first
-        argument of the model call function: :obj:`model(inputs)`. If you choose this second option, there are three
-        possibilities you can use to gather all the input Tensors in the first positional argument : - a single Tensor
-        with :obj:`input_ids` only and nothing else: :obj:`model(inputs_ids)` - a list of varying length with one or
-        several input Tensors IN THE ORDER given in the docstring: :obj:`model([input_ids, attention_mask])` or
-        :obj:`model([input_ids, attention_mask, token_type_ids])` - a dictionary with one or several input Tensors
-        associated to the input names given in the docstring: :obj:`model({"input_ids": input_ids, "token_type_ids":
-        token_type_ids})`
+        TF 2.0 models accepts two formats as inputs:
+
+        - having all inputs as keyword arguments (like PyTorch models), or
+        - having all inputs as a list, tuple or dict in the first positional arguments.
+
+        This second option is useful when using :meth:`tf.keras.Model.fit` method which currently requires having all
+        the tensors in the first argument of the model call function: :obj:`model(inputs)`.
+
+        If you choose this second option, there are three possibilities you can use to gather all the input Tensors in
+        the first positional argument :
+
+        - a single Tensor with :obj:`input_ids` only and nothing else: :obj:`model(inputs_ids)`
+        - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring:
+          :obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])`
+        - a dictionary with one or several input Tensors associated to the input names given in the docstring:
+          :obj:`model({"input_ids": input_ids, "token_type_ids": token_type_ids})`
 
     Parameters:
         config (:class:`~transformers.DPRConfig`): Model configuration class with all the parameters of the model.

src/transformers/models/layoutlm/modeling_layoutlm.py

@@ -638,8 +638,8 @@ LAYOUTLM_INPUTS_DOCSTRING = r"""
 
             `What are input IDs? <../glossary.html#input-ids>`__
         bbox (:obj:`torch.LongTensor` of shape :obj:`({0}, 4)`, `optional`):
-            Bounding Boxes of each input sequence tokens. Selected in the range ``[0, config.max_2d_position_embeddings
-            - 1]``.
+            Bounding Boxes of each input sequence tokens. Selected in the range ``[0,
+            config.max_2d_position_embeddings-1]``.
         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 MASKED tokens.

src/transformers/models/led/modeling_led.py

@@ -172,11 +172,11 @@ class LEDEncoderSelfAttention(nn.Module):
         :class:`LEDEncoderSelfAttention` expects `len(hidden_states)` to be multiple of `attention_window`. Padding to
         `attention_window` happens in :meth:`LEDEncoderModel.forward` to avoid redoing the padding on each layer.
 
-        The `attention_mask` is changed in :meth:`BertModel.forward` from 0, 1, 2 to -ve: no attention
-
-              0: local attention
-            +ve: global attention
+        The `attention_mask` is changed in :meth:`LEDEncoderModel.forward` from 0, 1, 2 to:
 
+            * -10000: no attention
+            * 0: local attention
+            * +10000: global attention
         """
         hidden_states = hidden_states.transpose(0, 1)
 

src/transformers/models/led/modeling_tf_led.py

@@ -190,11 +190,11 @@ class TFLEDEncoderSelfAttention(tf.keras.layers.Layer):
         LongformerSelfAttention expects `len(hidden_states)` to be multiple of `attention_window`. Padding to
         `attention_window` happens in LongformerModel.forward to avoid redoing the padding on each layer.
 
-        The `attention_mask` is changed in `BertModel.forward` from 0, 1, 2 to -ve: no attention
-
-              0: local attention
-            +ve: global attention
+        The `attention_mask` is changed in :meth:`LongformerModel.forward` from 0, 1, 2 to:
 
+            * -10000: no attention
+            * 0: local attention
+            * +10000: global attention
         """
         # retrieve input args
         (

src/transformers/models/longformer/modeling_longformer.py

@@ -561,11 +561,11 @@ class LongformerSelfAttention(nn.Module):
         :class:`LongformerSelfAttention` expects `len(hidden_states)` to be multiple of `attention_window`. Padding to
         `attention_window` happens in :meth:`LongformerModel.forward` to avoid redoing the padding on each layer.
 
-        The `attention_mask` is changed in :meth:`BertModel.forward` from 0, 1, 2 to -ve: no attention
-
-              0: local attention
-            +ve: global attention
+        The `attention_mask` is changed in :meth:`LongformerModel.forward` from 0, 1, 2 to:
 
+            * -10000: no attention
+            * 0: local attention
+            * +10000: global attention
         """
         hidden_states = hidden_states.transpose(0, 1)
 

src/transformers/models/longformer/modeling_tf_longformer.py

@@ -768,11 +768,11 @@ class TFLongformerSelfAttention(tf.keras.layers.Layer):
         LongformerSelfAttention expects `len(hidden_states)` to be multiple of `attention_window`. Padding to
         `attention_window` happens in LongformerModel.forward to avoid redoing the padding on each layer.
 
-        The `attention_mask` is changed in `BertModel.forward` from 0, 1, 2 to -ve: no attention
-
-              0: local attention
-            +ve: global attention
+        The `attention_mask` is changed in :meth:`LongformerModel.forward` from 0, 1, 2 to:
 
+            * -10000: no attention
+            * 0: local attention
+            * +10000: global attention
         """
         # retrieve input args
         (

src/transformers/testing_utils.py

@@ -522,6 +522,7 @@ class CaptureLogger:
     Context manager to capture `logging` streams
 
     Args:
+
     - logger: 'logging` logger object
 
     Results:
@@ -851,9 +852,10 @@ def pytest_terminal_summary_main(tr, id):
     there.
 
     Args:
+
     - tr: `terminalreporter` passed from `conftest.py`
-    - id: unique id like `tests` or `examples` that will be incorporated into the final reports
-      filenames - this is needed as some jobs have multiple runs of pytest, so we can't have them overwrite each other.
+    - id: unique id like `tests` or `examples` that will be incorporated into the final reports filenames - this is
+      needed as some jobs have multiple runs of pytest, so we can't have them overwrite each other.
 
     NB: this functions taps into a private _pytest API and while unlikely, it could break should
     pytest do internal changes - also it calls default internal methods of terminalreporter which

src/transformers/trainer_utils.py

@@ -191,6 +191,7 @@ def speed_metrics(split, start_time, num_samples=None):
     should be run immediately after the operation to be measured has completed.
 
     Args:
+
     - split: name to prefix metric (like train, eval, test...)
     - start_time: operation start time
     - num_samples: number of samples processed

utils/style_doc.py

@@ -42,7 +42,7 @@ DOC_SPECIAL_WORD = [
 # Matches any declaration of textual block, like `.. note::`. (ignore case to avoid writing all versions in the list)
 _re_textual_blocks = re.compile(r"^\s*\.\.\s+(" + "|".join(TEXTUAL_BLOCKS) + r")\s*::\s*$", re.IGNORECASE)
 # Matches list introduction in rst.
-_re_list = re.compile(r"^(\s*-\s+|\s*\*\s+|\s*\d+.\s+)")
+_re_list = re.compile(r"^(\s*-\s+|\s*\*\s+|\s*\d+\.\s+)")
 # Matches the indent in a line.
 _re_indent = re.compile(r"^(\s*)\S")
 # Matches a table declaration in rst.
@@ -355,10 +355,34 @@ rst_styler = CodeStyler()
 doc_styler = DocstringStyler()
 
 
+def _add_new_lines_before_list(text):
+    """Add a new empty line before a list begins."""
+    lines = text.split("\n")
+    new_lines = []
+    in_list = False
+    for idx, line in enumerate(lines):
+        # Detect if the line is the start of a new list.
+        if _re_list.search(line) is not None and not in_list:
+            current_indent = get_indent(line)
+            in_list = True
+            # If the line before is non empty, add an extra new line.
+            if idx > 0 and len(lines[idx - 1]) != 0:
+                new_lines.append("")
+        # Detect if we're out of the current list.
+        if in_list and not line.startswith(current_indent) and _re_list.search(line) is None:
+            in_list = False
+        new_lines.append(line)
+    return "\n".join(new_lines)
+
+
 def style_rst_file(doc_file, max_len=119, check_only=False):
     """ Style one rst file `doc_file` to `max_len`."""
     with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
         doc = f.read()
+
+    # Add missing new lines before lists
+    doc = _add_new_lines_before_list(doc)
+    # Style
     clean_doc = rst_styler.style(doc, max_len=max_len)
 
     diff = clean_doc != doc
@@ -391,6 +415,8 @@ def style_docstring(docstring, max_len=119):
 
     # Add missing new lines before Args/Returns etc.
     docstring = _re_any_doc_special_word.sub(r"\n\n\1\2\3\n", docstring)
+    # Add missing new lines before lists
+    docstring = _add_new_lines_before_list(docstring)
     # Style
     styled_doc = doc_styler.style(docstring, max_len=max_len, min_indent=indent)