Update modeling.py

modeling.py

@@ -355,6 +355,7 @@ class NewEmbeddings(nn.Module):
                     attention_mask[i, l:] = 0
 
         # Set attention_mask_bool for unpadding
+        unpad_inputs = False
         if unpad_inputs:
             attention_mask_bool = attention_mask.bool()
             if length is None:
@@ -861,7 +862,8 @@ class NewModel(NewPreTrainedModel):
             output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
         )
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-        unpad_inputs = unpad_inputs if unpad_inputs is not None else self.config.unpad_inputs
+        # unpad_inputs = unpad_inputs if unpad_inputs is not None else self.config.unpad_inputs
+        unpad_inputs = False
         output_padded = length is None
 
         if input_ids is not None and inputs_embeds is not None:
@@ -1006,7 +1008,7 @@ class NewForMaskedLM(NewPreTrainedModel):
         """
 
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
+        unpad_inputs = False
         if labels is None or not self.new.config.unpad_inputs:
             length = None
             subset_indices = None
@@ -1027,7 +1029,7 @@ class NewForMaskedLM(NewPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            unpad_inputs=False,
+            unpad_inputs=unpad_inputs,
         )
 
         sequence_output = outputs[0]
@@ -1092,7 +1094,7 @@ class NewForSequenceClassification(NewPreTrainedModel):
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
+        unpad_inputs = False
         outputs = self.new(
             input_ids,
             attention_mask=attention_mask,
@@ -1103,7 +1105,7 @@ class NewForSequenceClassification(NewPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            unpad_inputs=False,
+            unpad_inputs=unpad_inputs,
         )
 
         pooled_output = outputs[1]
@@ -1182,7 +1184,7 @@ class NewForMultipleChoice(NewPreTrainedModel):
         """
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
         num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]
-
+        unpad_inputs = False
         input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
         attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
         token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
@@ -1203,7 +1205,7 @@ class NewForMultipleChoice(NewPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            unpad_inputs=False,
+            unpad_inputs=unpad_inputs,
         )
 
         pooled_output = outputs[1]
@@ -1263,7 +1265,7 @@ class NewForTokenClassification(NewPreTrainedModel):
             Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
         """
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
+        unpad_inputs = False
         outputs = self.new(
             input_ids,
             attention_mask=attention_mask,
@@ -1274,7 +1276,7 @@ class NewForTokenClassification(NewPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            unpad_inputs=False,
+            unpad_inputs=unpad_inputs,
         )
 
         sequence_output = outputs[0]
@@ -1336,7 +1338,7 @@ class NewForQuestionAnswering(NewPreTrainedModel):
             are not taken into account for computing the loss.
         """
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
+        unpad_inputs = False
         outputs = self.new(
             input_ids,
             attention_mask=attention_mask,
@@ -1347,7 +1349,7 @@ class NewForQuestionAnswering(NewPreTrainedModel):
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            unpad_inputs=False,
+            unpad_inputs=unpad_inputs,
         )
 
         sequence_output = outputs[0]