Fix `input_embeds` docstring in encoder-decoder architectures (#28168)

src/transformers/models/seamless_m4t/modeling_seamless_m4t.py

@@ -1900,11 +1900,11 @@ class SeamlessM4TDecoder(SeamlessM4TPreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/seamless_m4t_v2/modeling_seamless_m4t_v2.py

@@ -2022,11 +2022,11 @@ class SeamlessM4Tv2Decoder(SeamlessM4Tv2PreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/speech_to_text/modeling_speech_to_text.py

@@ -933,11 +933,11 @@ class Speech2TextDecoder(Speech2TextPreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/speech_to_text/modeling_tf_speech_to_text.py

@@ -1064,11 +1064,11 @@ class TFSpeech2TextDecoder(tf.keras.layers.Layer):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`tf.Tensor` of shape
-                `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`tf.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/speech_to_text_2/modeling_speech_to_text_2.py

@@ -530,11 +530,11 @@ class Speech2Text2Decoder(Speech2Text2PreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/speecht5/modeling_speecht5.py

@@ -1569,11 +1569,11 @@ class SpeechT5Decoder(SpeechT5PreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/trocr/modeling_trocr.py

@@ -549,11 +549,11 @@ class TrOCRDecoder(TrOCRPreTrainedModel):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of
-                shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing
-                `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/whisper/modeling_tf_whisper.py

@@ -958,11 +958,11 @@ class TFWhisperDecoder(tf.keras.layers.Layer):
 
                 If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those
                 that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of
-                all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`tf.Tensor` of shape
-                `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `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.
+                all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+            inputs_embeds (`tf.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.

src/transformers/models/xglm/modeling_xglm.py

@@ -113,10 +113,11 @@ XGLM_INPUTS_DOCSTRING = r"""
 
             If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
             don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
-            `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape
-            `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `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.
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `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.
         output_attentions (`bool`, *optional*):
             Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
             tensors for more detail.

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

@@ -2083,8 +2083,11 @@ class {{cookiecutter.camelcase_modelname}}PreTrainedModel(PreTrainedModel):
 
             If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids`
             (those that don't have their past key value states given to this model) of shape `(batch_size, 1)`
-            instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `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.
+            instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `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.
         decoder_inputs_embeds (`torch.FloatTensor` of shape `(batch_size, target_sequence_length, hidden_size)`, *optional*):
             Optionally, instead of passing `decoder_input_ids` you can choose to directly pass an embedded
             representation. If `past_key_values` is used, optionally only the last `decoder_inputs_embeds`
@@ -2407,8 +2410,11 @@ class {{cookiecutter.camelcase_modelname}}Decoder({{cookiecutter.camelcase_model
                 If `past_key_values` are used, the user can optionally input only the last
                 `decoder_input_ids` (those that don't have their past key value states given to this model) of
                 shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape `(batch_size,
-                sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `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.
+                sequence_length)`.
+            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+                Optionally, instead of passing `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.
             output_attentions (`bool`, *optional*):
                 Whether or not to return the attentions tensors of all attention layers. See `attentions` under
                 returned tensors for more detail.