diff --git a/tensor2tensor/data_generators/all_problems.py b/tensor2tensor/data_generators/all_problems.py
index c1f54f18f..24d1fffed 100644
--- a/tensor2tensor/data_generators/all_problems.py
+++ b/tensor2tensor/data_generators/all_problems.py
@@ -19,6 +19,7 @@
 from __future__ import print_function
 
 import importlib
+from six.moves import range  # pylint: disable=redefined-builtin
 
 MODULES = [
     "tensor2tensor.data_generators.algorithmic",
@@ -97,7 +98,7 @@
 
 def _is_import_err_msg(err_str, module):
   parts = module.split(".")
-  suffixes = [".".join(parts[i:]) for i in xrange(len(parts))]
+  suffixes = [".".join(parts[i:]) for i in range(len(parts))]
   return err_str in (
       ["No module named %s" % suffix for suffix in suffixes] +
       ["No module named '%s'" % suffix for suffix in suffixes])
diff --git a/tensor2tensor/data_generators/generator_utils.py b/tensor2tensor/data_generators/generator_utils.py
index 8a722101a..aae3e5572 100644
--- a/tensor2tensor/data_generators/generator_utils.py
+++ b/tensor2tensor/data_generators/generator_utils.py
@@ -969,11 +969,11 @@ def random_deinterleave(text, separator_symbol="X"):
   cut = [False] * n
   cut[0] = True
   num_cuts = int(math.exp(random.uniform(0, math.log(n))))
-  for _ in xrange(num_cuts):
+  for _ in range(num_cuts):
     cut[random.randint(1, n -1)] = True
   out = [[], []]
   part = random.randint(0, 1)
-  for i in xrange(n):
+  for i in range(n):
     if cut[i]:
       out[part].append(separator_symbol)
       part = 1 - part
diff --git a/tensor2tensor/data_generators/transduction_problems.py b/tensor2tensor/data_generators/transduction_problems.py
index 6e1cff743..37082dcbf 100644
--- a/tensor2tensor/data_generators/transduction_problems.py
+++ b/tensor2tensor/data_generators/transduction_problems.py
@@ -33,7 +33,7 @@
 import os
 import random
 
-from six.moves import xrange  # pylint: disable=redefined-builtin
+from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensor2tensor.data_generators import problem
 from tensor2tensor.data_generators import text_encoder
@@ -123,7 +123,7 @@ def sequence_length(self, dataset_split):
                           self.max_sequence_length(dataset_split))
 
   def build_vocab(self):
-    return ["sym_%d" % i for i in xrange(1, self.num_symbols + 1)]
+    return ["sym_%d" % i for i in range(1, self.num_symbols + 1)]
 
   def get_or_create_vocab(self, data_dir, tmp_dir, force_get=False):
     vocab_filename = os.path.join(data_dir, self.vocab_filename)
@@ -144,7 +144,7 @@ def transpose_sequence(self, input_sequence):
     raise NotImplementedError()
 
   def generate_samples(self, data_dir, tmp_dir, dataset_split):
-    for _ in xrange(self.num_samples(dataset_split)):
+    for _ in range(self.num_samples(dataset_split)):
       source = self.generate_random_sequence(dataset_split)
       target = self.transpose_sequence(source)
       yield {
diff --git a/tensor2tensor/models/research/vqa_self_attention.py b/tensor2tensor/models/research/vqa_self_attention.py
index 4702b598f..c9ae04680 100644
--- a/tensor2tensor/models/research/vqa_self_attention.py
+++ b/tensor2tensor/models/research/vqa_self_attention.py
@@ -19,7 +19,7 @@
 from __future__ import division
 from __future__ import print_function
 
-from six.moves import xrange
+from six.moves import range  # pylint: disable=redefined-builtin
 
 from tensor2tensor.layers import common_attention
 from tensor2tensor.layers import common_hparams
@@ -657,7 +657,7 @@ def iterative_encoder_decoder(encoder_input,
                               query,
                               hparams):
   """Iterative encoder decoder."""
-  for _ in xrange(hparams.num_rec_steps):
+  for _ in range(hparams.num_rec_steps):
     with tf.variable_scope("step", reuse=tf.AUTO_REUSE):
       encoder_output = image_question_encoder(
           encoder_input,