diff --git a/test/models/test_bert.py b/test/models/test_bert.py
index 78e3367339..e2d7fad730 100644
--- a/test/models/test_bert.py
+++ b/test/models/test_bert.py
@@ -1,4 +1,6 @@
 #!/usr/bin/env python
+import os
+os.environ['USE_TF'] = '0'  # prevent transformers from importing tensorflow
 import unittest
 from tinygrad import Tensor
 import numpy as np
diff --git a/test/models/test_efficientnet.py b/test/models/test_efficientnet.py
index df936299d2..62ad994f4d 100644
--- a/test/models/test_efficientnet.py
+++ b/test/models/test_efficientnet.py
@@ -68,7 +68,7 @@ class TestEfficientNet(unittest.TestCase):
     self.assertEqual(_LABELS[labels[0]], "sports car, sport car")
 
   def test_chicken_car(self):
-    labels = _infer(self.model, np.concat([chicken_img, car_img], axis=0))
+    labels = _infer(self.model, np.concatenate([chicken_img, car_img], axis=0))
     self.assertEqual(_LABELS[labels[0]], "hen")
     self.assertEqual(_LABELS[labels[1]], "sports car, sport car")
 
diff --git a/test/unit/test_hashing.py b/test/unit/test_hashing.py
index cbee0f6bfb..c3876786dc 100644
--- a/test/unit/test_hashing.py
+++ b/test/unit/test_hashing.py
@@ -47,10 +47,10 @@ class TestKeccak(unittest.TestCase):
 
       ha_ref, hb_ref = hasher(a), hasher(b)
       tres = Tensor.stack(*(Tensor(d) for d in (a, b))).keccak(name)
-      ha, hb = tres[0].data(), tres[1].data()
+      ha, hb = bytes(tres[0].data()), bytes(tres[1].data())
 
       self.assertEqual(ha_ref, ha)
-      self.assertEqual(ha_ref, Tensor(a).keccak(name).data())
+      self.assertEqual(ha_ref, bytes(Tensor(a).keccak(name).data()))
       self.assertEqual(hb_ref, hb)
 
   def test_referenced(self):
diff --git a/tinygrad/tensor.py b/tinygrad/tensor.py
index 290a36e568..447eba179a 100644
--- a/tinygrad/tensor.py
+++ b/tinygrad/tensor.py
@@ -1845,6 +1845,7 @@ class Tensor(OpMixin):
         p = state.reshape(bs, 5, 5).transpose(2, 1)
         t1 = (p[:,:,0] ^ p[:,:,1] ^ p[:,:,2] ^ p[:,:,3] ^ p[:,:,4]).roll(-1, 1) # xor reduce
         state = state ^ (t1.roll(2, 1).bitwise_xor((t1 << 1) ^ (t1 >> 63)).unsqueeze(2).expand(bs, 5, 5).transpose(2, 1).flatten(1))
+        state = state.contiguous()  # required for correct indexing in π step # TODO: why is it needed?
         # ρ and π steps
         state = state[:, reorder_indexes]
         state = (state * rot_offsets_v0).bitwise_or(state // rot_offsets_v1).reshape(bs, 5, 5)