Upload 7 files

modeling_indictrans.py

@@ -61,7 +61,8 @@ def shift_tokens_right(input_ids: torch.Tensor, pad_token_id: int, decoder_start
 
 
 def prepare_decoder_input_ids_label(decoder_input_ids, decoder_attention_mask):
-    labels = decoder_input_ids[:, 1:]
+    labels = decoder_input_ids.full_size(decoder_input_ids.size(), -100)
+    labels[:, :-1] = decoder_input_ids[:, 1:]
 
     labels_mask = labels == 1
     labels[labels_mask] = -100
@@ -70,6 +71,8 @@ def prepare_decoder_input_ids_label(decoder_input_ids, decoder_attention_mask):
     decoder_input_ids[mask] = 1
     decoder_attention_mask[mask] = 0
 
+    labels = decoder_input_ids[:, 1:]
+
     return decoder_input_ids, decoder_attention_mask, labels
 
 
@@ -1247,7 +1250,7 @@ class IndicTransForConditionalGeneration(IndicTransPreTrainedModel):
             # move labels to the correct device to enable PP
             labels = labels.to(lm_logits.device)
             loss_fct = nn.CrossEntropyLoss()
-            masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.decoder_vocab_size), labels.view(-1))
+            masked_lm_loss = loss_fct(lm_logits.view(-1, self.config.decoder_vocab_size), labels.reshape(-1))
 
         if not return_dict:
             output = (lm_logits,) + outputs[1:]