Remove device parameter from create_extended_attention_mask_for_decoder (#16894)

src/transformers/models/roberta/modeling_roberta.py

@@ -817,7 +817,7 @@ class RobertaModel(RobertaPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/roformer/modeling_roformer.py

@@ -900,7 +900,7 @@ class RoFormerModel(RoFormerPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/splinter/modeling_splinter.py

@@ -710,7 +710,7 @@ class SplinterModel(SplinterPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/squeezebert/modeling_squeezebert.py

@@ -612,7 +612,7 @@ class SqueezeBertModel(SqueezeBertPreTrainedModel):
         if token_type_ids is None:
             token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=device)
 
-        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape)
         # Prepare head mask if needed
         # 1.0 in head_mask indicate we keep the head
         # attention_probs has shape bsz x n_heads x N x N

src/transformers/models/t5/modeling_t5.py

@@ -957,7 +957,7 @@ class T5Stack(T5PreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape, inputs_embeds.device)
+        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/tapas/modeling_tapas.py

@@ -954,7 +954,7 @@ class TapasModel(TapasPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D ou 3D attention mask is provided for the cross-attention
         # we need to make broadcastabe to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/vilt/modeling_vilt.py

@@ -843,7 +843,7 @@ class ViltModel(ViltPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         encoder_outputs = self.encoder(
             embedding_output,

src/transformers/models/visual_bert/modeling_visual_bert.py

@@ -794,12 +794,12 @@ class VisualBertModel(VisualBertPreTrainedModel):
         if visual_embeds is not None:
             combined_attention_mask = torch.cat((attention_mask, visual_attention_mask), dim=-1)
             extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(
-                combined_attention_mask, [batch_size, input_shape + visual_input_shape], device
+                combined_attention_mask, (batch_size, input_shape + visual_input_shape)
             )
 
         else:
             extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(
-                attention_mask, [batch_size, input_shape], device
+                attention_mask, (batch_size, input_shape)
             )
 
         # Prepare head mask if needed

src/transformers/models/xlm_roberta_xl/modeling_xlm_roberta_xl.py

@@ -788,7 +788,7 @@ class XLMRobertaXLModel(XLMRobertaXLPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]

src/transformers/models/yoso/modeling_yoso.py

@@ -816,7 +816,7 @@ class YosoModel(YosoPreTrainedModel):
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # Prepare head mask if needed
         # 1.0 in head_mask indicate we keep the head

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

@@ -876,7 +876,7 @@ class {{cookiecutter.camelcase_modelname}}Model({{cookiecutter.camelcase_modelna
 
         # We can provide a self-attention mask of dimensions [batch_size, from_seq_length, to_seq_length]
         # ourselves in which case we just need to make it broadcastable to all heads.
-        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape, device)
+        extended_attention_mask: torch.Tensor = self.get_extended_attention_mask(attention_mask, input_shape)
 
         # If a 2D or 3D attention mask is provided for the cross-attention
         # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]