diff --git a/tensorflow/python/BUILD b/tensorflow/python/BUILD
index ce1b525f2f7..8be4c143ab8 100644
--- a/tensorflow/python/BUILD
+++ b/tensorflow/python/BUILD
@@ -4885,7 +4885,6 @@ cuda_py_test(
         ":test_ops",
         ":unconnected_gradients",
         ":variable_scope",
-        "//tensorflow/python/keras:engine",
         "//third_party/py/numpy",
         "@absl_py//absl/testing:parameterized",
     ],
diff --git a/tensorflow/python/keras/integration_test/BUILD b/tensorflow/python/keras/integration_test/BUILD
index d3920224b58..d40006d161a 100644
--- a/tensorflow/python/keras/integration_test/BUILD
+++ b/tensorflow/python/keras/integration_test/BUILD
@@ -32,6 +32,16 @@ tf_py_test(
     ],
 )
 
+tf_py_test(
+    name = "gradients_test",
+    srcs = ["gradients_test.py"],
+    python_version = "PY3",
+    deps = [
+        "//tensorflow:tensorflow_py",
+        "//tensorflow/python:extra_py_tests_deps",
+    ],
+)
+
 tf_py_test(
     name = "legacy_rnn_test",  # Remove this target in when TF 1 is deprecated.
     srcs = ["legacy_rnn_test.py"],
diff --git a/tensorflow/python/keras/integration_test/gradients_test.py b/tensorflow/python/keras/integration_test/gradients_test.py
new file mode 100644
index 00000000000..706150ddaa8
--- /dev/null
+++ b/tensorflow/python/keras/integration_test/gradients_test.py
@@ -0,0 +1,84 @@
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import tensorflow as tf
+
+
+class TestKerasModelClass(tf.keras.Model):
+  """A simple tensorflow keras Model class definition."""
+
+  def __init__(self, width):
+    super(TestKerasModelClass, self).__init__()
+    self.width = width
+
+  def build(self, input_shape):
+    self.weight = self.add_weight(
+        name="test_keras_var",
+        shape=(self.width,),
+        dtype=tf.float32,
+        trainable=True,
+    )
+
+  def call(self, inputs):
+    return self.weight * inputs
+
+
+class GradientsTest(tf.test.TestCase):
+
+  def _TestVariablesGradient(self, inputs, test_model, vars_to_grad):
+    """Returns gradients of `test_model` with respect to `vars_to_grad`."""
+
+    test_model_re = tf.recompute_grad(test_model)
+
+    with tf.GradientTape(persistent=True) as tape:
+      tape.watch(vars_to_grad)
+      out_re = test_model_re(inputs)
+      out = test_model(inputs)
+
+    grads_re = tape.gradient(out_re, vars_to_grad)
+    grads = tape.gradient(out, vars_to_grad)
+
+    return grads_re, grads
+
+  def testKerasRecompute(self):
+    """Checks that recompute_grad works for a simple Keras Model."""
+
+    test_model = TestKerasModelClass(10)
+    test_input = tf.constant(tf.zeros((10, 10), dtype=np.float32))
+    # Ensures keras model is initialized.
+    test_model(test_input)  # pylint: disable=not-callable
+    grads_re, grads = self._TestVariablesGradient(test_input, test_model,
+                                                  test_input)
+
+    grads_re = self.evaluate(grads_re)
+    grads = self.evaluate(grads)
+    for g, g_re in zip(grads, grads_re):
+      self.assertAllClose(g, g_re)
+
+    grads_re, grads = self._TestVariablesGradient(test_input, test_model,
+                                                  test_model.variables)
+
+    grads_re = self.evaluate(grads_re)
+    grads = self.evaluate(grads)
+    for g, g_re in zip(grads, grads_re):
+      self.assertAllClose(g, g_re)
+
+
+if __name__ == "__main__":
+  tf.test.main()
diff --git a/tensorflow/python/ops/gradients_test.py b/tensorflow/python/ops/gradients_test.py
index 9e9ebccfe0f..817d8a1adbe 100644
--- a/tensorflow/python/ops/gradients_test.py
+++ b/tensorflow/python/ops/gradients_test.py
@@ -33,7 +33,6 @@ from tensorflow.python.framework import ops
 from tensorflow.python.framework import test_ops
 from tensorflow.python.framework import test_util
 from tensorflow.python.framework.constant_op import constant
-from tensorflow.python.keras.engine import training
 from tensorflow.python.layers import core as core_layers
 from tensorflow.python.ops import array_grad  # pylint: disable=unused-import
 from tensorflow.python.ops import array_ops
@@ -1324,41 +1323,8 @@ class TensorListGradientsTest(test_util.TensorFlowTestCase):
         self.assertEquals(self.evaluate(grad), 5.)
 
 
-class TestKerasModelClass(training.Model):
-  """A simple tensorflow keras Model class definition."""
-
-  def __init__(self, width):
-    super(TestKerasModelClass, self).__init__()
-
-    self.weight = variable_scope.get_variable(
-        name="test_keras_var",
-        shape=width,
-        dtype=dtypes.float32,
-        trainable=True,
-        use_resource=True,
-    )
-
-  def call(self, inputs):
-    return self.weight * inputs
-
-
 class VariablesGradientTest(test_util.TensorFlowTestCase):
 
-  def _TestVariablesGradient(self, inputs, test_model, vars_to_grad):
-    """Returns gradients of `test_model` with respect to `vars_to_grad`."""
-
-    test_model_re = custom_gradient.recompute_grad(test_model)
-
-    with backprop.GradientTape(persistent=True) as tape:
-      tape.watch(vars_to_grad)
-      out_re = test_model_re(inputs)
-      out = test_model(inputs)
-
-    grads_re = tape.gradient(out_re, vars_to_grad)
-    grads = tape.gradient(out, vars_to_grad)
-
-    return grads_re, grads
-
   def _TestFnVariablesGradient(self, inputs, test_fn, vars_to_grad):
     """Returns gradients of `test_model` with respect to `vars_to_grad`."""
 
@@ -1374,30 +1340,6 @@ class VariablesGradientTest(test_util.TensorFlowTestCase):
 
     return grads_re, grads
 
-  @test_util.run_in_graph_and_eager_modes
-  def testKerasRecompute(self):
-    """Checks that recompute_grad works for a simple Keras Model."""
-
-    test_model = TestKerasModelClass(10)
-    test_input = constant(np.zeros((10, 10), dtype=np.float32))
-    self.evaluate(variables.global_variables_initializer())
-    test_model(test_input)  # Ensures keras model is initialized.
-    grads_re, grads = self._TestVariablesGradient(test_input, test_model,
-                                                  test_input)
-
-    grads_re = self.evaluate(grads_re)
-    grads = self.evaluate(grads)
-    for g, g_re in zip(grads, grads_re):
-      self.assertAllClose(g, g_re)
-
-    grads_re, grads = self._TestVariablesGradient(test_input, test_model,
-                                                  test_model.variables)
-
-    grads_re = self.evaluate(grads_re)
-    grads = self.evaluate(grads)
-    for g, g_re in zip(grads, grads_re):
-      self.assertAllClose(g, g_re)
-
   @test_util.run_in_graph_and_eager_modes
   def testFnRecompute(self):
     """Checks that recompute_grad works grads of function args."""