diff --git a/attacut/dataloaders.py b/attacut/dataloaders.py
index 21c3f07..a84c72a 100644
--- a/attacut/dataloaders.py
+++ b/attacut/dataloaders.py
@@ -175,7 +175,7 @@ def _process_line(line):
     def collate_fn(batch):
         total_samples = len(batch)
 
-        seq_lengths = np.array(list(map(lambda x: x[0][1], batch)))
+        seq_lengths = np.array(list(map(lambda x: x[0][1], batch)), dtype=np.int64)
         max_length = np.max(seq_lengths)
 
         features = np.zeros((total_samples, 2, max_length), dtype=np.int64)
diff --git a/scripts/attacut-cli b/scripts/attacut-cli
index f4a5a34..a75e171 100755
--- a/scripts/attacut-cli
+++ b/scripts/attacut-cli
@@ -165,8 +165,10 @@ if __name__ == "__main__":
                   pred = preds[after_sorting_ix, :]
                   token = tokens[ori_ix]
 
-
-                  words = preprocessing.find_words_from_preds(token, pred)
+                  if len(token) != 0:
+                      words = preprocessing.find_words_from_preds(token, pred)
+                  else:
+                      words = ""
                   fout.write("%s\n" % SEP.join(words))
 
               tq.update(n=preds.shape[0])
\ No newline at end of file