diff --git a/LLaVA/llava/train/train.py b/LLaVA/llava/train/train.py
index a868800..4f4a121 100644
--- a/LLaVA/llava/train/train.py
+++ b/LLaVA/llava/train/train.py
@@ -781,16 +781,18 @@ def make_supervised_data_module(tokenizer: transformers.PreTrainedTokenizer,
                                 data_args=data_args)
     data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
 
-    eval_dataset = None
     validation_data_path = data_args.data_path.replace('train', 'val')
-    if os.path.isfile(validation_data_path):
-        eval_dataset = LazySupervisedDataset(tokenizer=tokenizer,
-                                    data_path=validation_data_path,
-                                    data_args=data_args)
-        print('Eval dataset found at {}'.format(validation_data_path))
-    else:
-        print('No validation dataset found at {}'.format(validation_data_path))
-        exit(1)
+
+    assert os.path.isfile(validation_data_path), (
+        'Validation file not found at {}.'.format(validation_data_path) + 
+        'Make sure you have run preprocess.py first.'
+    )
+
+    eval_dataset = LazySupervisedDataset(tokenizer=tokenizer,
+                                data_path=validation_data_path,
+                                data_args=data_args)
+    print('Eval dataset found at {}'.format(validation_data_path))
+
     return dict(train_dataset=train_dataset,
                 eval_dataset=eval_dataset,
                 data_collator=data_collator)