Update missing docs on activation_dropout and fix DropOut docs for SEW-D (#26031)

* add missing doc for activation dropout

* fix doc for SEW-D dropout

* deprecate hidden_dropout for SEW-D

src/transformers/models/data2vec/configuration_data2vec_audio.py

@@ -58,6 +58,8 @@ class Data2VecAudioConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/hubert/configuration_hubert.py

@@ -58,6 +58,8 @@ class HubertConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout(`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout(`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/sew/configuration_sew.py

@@ -59,6 +59,8 @@ class SEWConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/sew_d/configuration_sew_d.py

@@ -72,6 +72,8 @@ class SEWDConfig(PretrainedConfig):
             The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
             `"relu"`, `"selu"`, `"gelu_python"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
+            Deprecated. Not used by the model and will be removed in a future version.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
@@ -238,7 +240,7 @@ class SEWDConfig(PretrainedConfig):
         self.pos_att_type = list(pos_att_type)
         self.hidden_act = hidden_act
         self.num_attention_heads = num_attention_heads
-        self.hidden_dropout = hidden_dropout
+        self._hidden_dropout = hidden_dropout
         self.attention_dropout = attention_dropout
         self.activation_dropout = activation_dropout
         self.feat_proj_dropout = feat_proj_dropout
@@ -280,3 +282,16 @@ class SEWDConfig(PretrainedConfig):
     @property
     def inputs_to_logits_ratio(self):
         return functools.reduce(operator.mul, self.conv_stride, 1)
+
+    @property
+    def hidden_dropout(self):
+        logger.warning_once("hidden_dropout is not used by the model and will be removed as config attribute in v4.35")
+        return self._hidden_dropout
+
+    def to_dict(self):
+        """
+        Serializes this instance to a Python dictionary.
+        """
+        output = super().to_dict()
+        output["hidden_dropout"] = output.pop("_hidden_dropout")
+        return output

src/transformers/models/unispeech/configuration_unispeech.py

@@ -61,6 +61,8 @@ class UniSpeechConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/unispeech_sat/configuration_unispeech_sat.py

@@ -62,6 +62,8 @@ class UniSpeechSatConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/wav2vec2/configuration_wav2vec2.py

@@ -59,6 +59,8 @@ class Wav2Vec2Config(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/wav2vec2_conformer/configuration_wav2vec2_conformer.py

@@ -61,6 +61,8 @@ class Wav2Vec2ConformerConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):

src/transformers/models/wavlm/configuration_wavlm.py

@@ -58,6 +58,8 @@ class WavLMConfig(PretrainedConfig):
             `"relu"`, `"selu"` and `"gelu_new"` are supported.
         hidden_dropout (`float`, *optional*, defaults to 0.1):
             The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        activation_dropout (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for activations inside the fully connected layer.
         attention_dropout (`float`, *optional*, defaults to 0.1):
             The dropout ratio for the attention probabilities.
         final_dropout (`float`, *optional*, defaults to 0.1):