Remove a TF usage warning and rework the documentation (#9756)

* Rework documentation

* Update the template

* Trigger CI

* Restore the warning but with the TF logger

* Update convbert doc

src/transformers/models/mpnet/modeling_tf_mpnet.py

@@ -748,12 +748,15 @@ MPNET_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/openai/modeling_tf_openai.py

@@ -485,12 +485,15 @@ OPENAI_GPT_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/pegasus/modeling_tf_pegasus.py

@@ -598,14 +598,20 @@ PEGASUS_INPUTS_DOCSTRING = r"""
         use_cache (:obj:`bool`, `optional`, defaults to :obj:`True`):
             If set to :obj:`True`, :obj:`past_key_values` key value states are returned and can be used to speed up
             decoding (see :obj:`past_key_values`). Set to :obj:`False` during training, :obj:`True` during generation
+            output_attentions (:obj:`bool`, `optional`): Whether or not to return the attentions tensors of all
+            attention layers. See ``attentions`` under returned tensors for more detail. This argument can be used only
+            in eager mode, in graph mode the value in the config will be used instead.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).
@@ -689,12 +695,18 @@ class TFPegasusEncoder(tf.keras.layers.Layer):
                 into associated vectors than the model's internal embedding lookup matrix.
             output_attentions (:obj:`bool`, `optional`):
                 Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under
-                returned tensors for more detail.
+                returned tensors for more detail. This argument can be used only in eager mode, in graph mode the value
+                in the config will be used instead.
             output_hidden_states (:obj:`bool`, `optional`):
                 Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors
-                for more detail.
+                for more detail. This argument can be used only in eager mode, in graph mode the value in the config
+                will be used instead.
             return_dict (:obj:`bool`, `optional`):
-                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+                argument can be used in eager mode, in graph mode the value will always be set to True.
+            training (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not to use the model in training mode (some modules like dropout modules have different
+                behaviors between training and evaluation).
         """
         inputs = input_processing(
             func=self.call,
@@ -881,12 +893,18 @@ class TFPegasusDecoder(tf.keras.layers.Layer):
                 into associated vectors than the model's internal embedding lookup matrix.
             output_attentions (:obj:`bool`, `optional`):
                 Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under
-                returned tensors for more detail.
+                returned tensors for more detail. This argument can be used only in eager mode, in graph mode the value
+                in the config will be used instead.
             output_hidden_states (:obj:`bool`, `optional`):
                 Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors
-                for more detail.
+                for more detail. This argument can be used only in eager mode, in graph mode the value in the config
+                will be used instead.
             return_dict (:obj:`bool`, `optional`):
-                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+                argument can be used in eager mode, in graph mode the value will always be set to True.
+            training (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not to use the model in training mode (some modules like dropout modules have different
+                behaviors between training and evaluation).
         """
         inputs = input_processing(
             func=self.call,

src/transformers/models/roberta/modeling_tf_roberta.py

@@ -784,12 +784,15 @@ ROBERTA_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/t5/modeling_tf_t5.py

@@ -999,12 +999,15 @@ T5_INPUTS_DOCSTRING = r"""
             decoding (see :obj:`past_key_values`).
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/transfo_xl/modeling_tf_transfo_xl.py

@@ -828,12 +828,15 @@ TRANSFO_XL_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/xlm/modeling_tf_xlm.py

@@ -669,12 +669,15 @@ XLM_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

src/transformers/models/xlnet/modeling_tf_xlnet.py

@@ -1133,12 +1133,15 @@ XLNET_INPUTS_DOCSTRING = r"""
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).

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

@@ -778,12 +778,15 @@ class TF{{cookiecutter.camelcase_modelname}}PreTrainedModel(TFPreTrainedModel):
             vectors than the model's internal embedding lookup matrix.
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).
@@ -2016,12 +2019,15 @@ class TF{{cookiecutter.camelcase_modelname}}PreTrainedModel(TFPreTrainedModel):
             decoding (see :obj:`past_key_values`). Set to :obj:`False` during training, :obj:`True` during generation
         output_attentions (:obj:`bool`, `optional`):
             Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned
-            tensors for more detail.
+            tensors for more detail. This argument can be used only in eager mode, in graph mode the value in the
+            config will be used instead.
         output_hidden_states (:obj:`bool`, `optional`):
             Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for
-            more detail.
+            more detail. This argument can be used only in eager mode, in graph mode the value in the config will be
+            used instead.
         return_dict (:obj:`bool`, `optional`):
-            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+            Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+            argument can be used in eager mode, in graph mode the value will always be set to True.
         training (:obj:`bool`, `optional`, defaults to :obj:`False`):
             Whether or not to use the model in training mode (some modules like dropout modules have different
             behaviors between training and evaluation).
@@ -2100,12 +2106,18 @@ class TF{{cookiecutter.camelcase_modelname}}Encoder(tf.keras.layers.Layer):
                 into associated vectors than the model's internal embedding lookup matrix.
             output_attentions (:obj:`bool`, `optional`):
                 Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under
-                returned tensors for more detail.
+                returned tensors for more detail. This argument can be used only in eager mode, in graph mode the value
+                in the config will be used instead.
             output_hidden_states (:obj:`bool`, `optional`):
                 Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors
-                for more detail.
+                for more detail. This argument can be used only in eager mode, in graph mode the value in the config
+                will be used instead.
             return_dict (:obj:`bool`, `optional`):
-                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+                argument can be used in eager mode, in graph mode the value will always be set to True.
+            training (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not to use the model in training mode (some modules like dropout modules have different
+                behaviors between training and evaluation).
         """
         inputs = input_processing(
             func=self.call,
@@ -2263,12 +2275,18 @@ class TF{{cookiecutter.camelcase_modelname}}Decoder(tf.keras.layers.Layer):
                 into associated vectors than the model's internal embedding lookup matrix.
             output_attentions (:obj:`bool`, `optional`):
                 Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under
-                returned tensors for more detail.
+                returned tensors for more detail. This argument can be used only in eager mode, in graph mode the value
+                in the config will be used instead.
             output_hidden_states (:obj:`bool`, `optional`):
                 Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors
-                for more detail.
+                for more detail. This argument can be used only in eager mode, in graph mode the value in the config
+                will be used instead.
             return_dict (:obj:`bool`, `optional`):
-                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple.
+                Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. This
+                argument can be used in eager mode, in graph mode the value will always be set to True.
+            training (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Whether or not to use the model in training mode (some modules like dropout modules have different
+                behaviors between training and evaluation).
         """
         inputs = input_processing(
             func=self.call,