fix xlnet bug (#1937)

Author: FrostML

examples/language_model/xlnet/run_glue.py

@@ -201,6 +201,8 @@ def create_data_loader(args, tokenizer):
             collate_fn=batchify_fn,
             num_workers=0,
             return_list=True)
+
+        return train_data_loader, dev_data_loader_matched, dev_data_loader_mismatched, train_ds, dev_ds_matched, dev_ds_mismatched
     else:
         dev_ds = load_dataset('glue', args.task_name, splits='dev')
         dev_ds = dev_ds.map(trans_func, lazy=True)
@@ -214,7 +216,7 @@ def create_data_loader(args, tokenizer):
             num_workers=0,
             return_list=True)
 
-    return train_data_loader, dev_data_loader, train_ds, dev_ds
+        return train_data_loader, dev_data_loader, train_ds, dev_ds
 
 
 def do_train(args):
@@ -231,8 +233,12 @@ def do_train(args):
 
     tokenizer = tokenizer_class.from_pretrained(args.model_name_or_path)
 
-    train_data_loader, dev_data_loader, train_ds, dev_ds = create_data_loader(
-        args, tokenizer)
+    if args.task_name == "mnli":
+        train_data_loader, dev_data_loader_matched, dev_data_loader_mismatched, train_ds, dev_ds_matched, dev_ds_mismatched = create_data_loader(
+            args, tokenizer)
+    else:
+        train_data_loader, dev_data_loader, train_ds, dev_ds = create_data_loader(
+            args, tokenizer)
 
     num_classes = 1 if train_ds.label_list is None else len(train_ds.label_list)
     model = XLNetForSequenceClassification.from_pretrained(

tests/test_tipc/benchmark/modules/xlnet.py

@@ -44,13 +44,21 @@ def create_data_loader(self, args, **kwargs):
         args.task_name = args.task_name.lower()
         tokenizer = XLNetTokenizer.from_pretrained(args.model_name_or_path)
 
-        train_loader, dev_loader, train_ds, dev_ds = create_data_loader(
-            args, tokenizer)
+        if args.task_name == "mnli":
+            train_data_loader, dev_data_loader_matched, dev_data_loader_mismatched, train_ds, _, _ = create_data_loader(
+                args, tokenizer)
+        else:
+            train_loader, dev_loader, train_ds, _ = create_data_loader(
+                args, tokenizer)
 
         self.num_batch = len(train_loader)
         self.label_list = train_ds.label_list
 
-        return train_loader, dev_loader
+        if args.task_name == "mnli":
+            return train_data_loader, (dev_data_loader_matched,
+                                       dev_data_loader_mismatched)
+        else:
+            return train_loader, dev_loader
 
     def build_model(self, args, **kwargs):
         num_classes = 1 if self.label_list is None else len(self.label_list)