ignore SQuAD targets outside of seq_length

modeling.py

@@ -455,9 +455,15 @@ class BertForQuestionAnswering(nn.Module):
         end_logits = end_logits.squeeze(-1)
 
         if start_positions is not None and end_positions is not None:
-            start_positions = start_positions.squeeze(-1) # If we are on multi-GPU, split add a dimension
+            # If we are on multi-GPU, split add a dimension - if not this is a no-op
+            start_positions = start_positions.squeeze(-1)
             end_positions = end_positions.squeeze(-1)
-            loss_fct = CrossEntropyLoss()
+            # sometimes the start/end positions are outside our model inputs, we ignore these terms
+            ignored_index = start_logits.size(1) + 1
+            start_positions.clamp_(0, ignored_index)
+            end_positions.clamp_(0, ignored_index)
+
+            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
             start_loss = loss_fct(start_logits, start_positions)
             end_loss = loss_fct(end_logits, end_positions)
             total_loss = (start_loss + end_loss) / 2