Merge pull request tensorflow#37919 from reedwm/none_grad_fix

goldiegadde · web-flow · commit e6e5d6df2ab2 · 2020-03-26T15:19:19.000-07:00
2.2-rc2 cherry-pick request: Fix crash in Model.fit() if a gradient is None
diff --git a/tensorflow/python/keras/distribute/distribute_strategy_test.py b/tensorflow/python/keras/distribute/distribute_strategy_test.py
@@ -804,6 +804,33 @@ def test_predict_multi_output_model_with_partial_batch(
           atol=1e-4,
           rtol=1e-4)
 
+  @combinations.generate(all_strategy_combinations_plus_run_distributed())
+  def test_gradients_are_none(self, distribution):
+
+    if not context.executing_eagerly():
+      self.skipTest('None gradients are not supported in graph mode')
+
+    class DenseWithExtraWeight(keras.layers.Dense):
+
+      def build(self, input_shape):
+        # Gradients w.r.t. extra_weights are None
+        self.extra_weight_1 = self.add_weight('extra_weight_1', shape=(),
+                                              initializer='ones')
+        super(DenseWithExtraWeight, self).build(input_shape)
+        self.extra_weight_2 = self.add_weight('extra_weight_2', shape=(),
+                                              initializer='ones')
+
+    with distribution.scope():
+      model = keras.Sequential([DenseWithExtraWeight(4, input_shape=(4,))])
+      model.compile('adam', 'mse')
+
+    inputs = np.random.normal(size=(64, 4))
+    targets = np.random.normal(size=(64, 4))
+    old_kernel = model.get_weights()[1]
+    model.fit(inputs, targets)
+    new_kernel = model.get_weights()[1]
+    self.assertNotAllEqual(old_kernel, new_kernel)
+
 
 class TestDistributionStrategyWithDatasets(test.TestCase,
                                            parameterized.TestCase):
diff --git a/tensorflow/python/keras/engine/training_test.py b/tensorflow/python/keras/engine/training_test.py
@@ -1364,6 +1364,30 @@ def apply_gradients(self, grads_and_vars, name=None):  # pylint: disable=useless
     model.fit(x, y)
     self.assertEqual(model.optimizer.aggregate_gradients_called, True)
 
+  @keras_parameterized.run_all_keras_modes(always_skip_v1=True)
+  def test_gradients_are_none(self):
+
+    class DenseWithExtraWeight(keras.layers.Dense):
+
+      def build(self, input_shape):
+        # Gradients w.r.t. extra_weights are None
+        self.extra_weight_1 = self.add_weight('extra_weight_1', shape=(),
+                                              initializer='ones')
+        super(DenseWithExtraWeight, self).build(input_shape)
+        self.extra_weight_2 = self.add_weight('extra_weight_2', shape=(),
+                                              initializer='ones')
+
+    model = keras.models.Sequential([DenseWithExtraWeight(4, input_shape=(4,))])
+    # Test clipping can handle None gradients
+    opt = keras.optimizer_v2.adam.Adam(clipnorm=1.0, clipvalue=1.0)
+    model.compile(opt, 'mse', run_eagerly=testing_utils.should_run_eagerly())
+    inputs = np.random.normal(size=(64, 4))
+    targets = np.random.normal(size=(64, 4))
+    old_kernel = model.get_weights()[1]
+    model.fit(inputs, targets)
+    new_kernel = model.get_weights()[1]
+    self.assertNotAllEqual(old_kernel, new_kernel)
+
 
 class TestExceptionsAndWarnings(keras_parameterized.TestCase):
 
diff --git a/tensorflow/python/keras/optimizer_v2/optimizer_v2.py b/tensorflow/python/keras/optimizer_v2/optimizer_v2.py
@@ -344,15 +344,16 @@ def _clip_gradients(self, grads):
         raise ValueError("Gradient clipping in the optimizer "
                          "(by setting clipnorm or clipvalue) is currently "
                          "unsupported when using a distribution strategy.")
-      grads = [clip_ops.clip_by_norm(g, self.clipnorm) for g in grads]
+      grads = [None if g is None else clip_ops.clip_by_norm(g, self.clipnorm)
+               for g in grads]
     if self.clipvalue is not None:
       if distribute_ctx.has_strategy():
         raise ValueError("Gradient clipping in the optimizer "
                          "(by setting clipnorm or clipvalue) is currently "
                          "unsupported when using a distribution strategy.")
+      v = self.clipvalue
       grads = [
-          clip_ops.clip_by_value(g, -self.clipvalue, self.clipvalue)
-          for g in grads
+          None if g is None else clip_ops.clip_by_value(g, -v, v) for g in grads
       ]
     return grads
 
@@ -521,6 +522,7 @@ def _aggregate_gradients(self, grads_and_vars):
       A list of all-reduced gradients.
     """
     grads_and_vars = list(grads_and_vars)
+    filtered_grads_and_vars = _filter_grads(grads_and_vars)
     def all_reduce_fn(distribution, grads_and_vars):
       return distribution.extended.batch_reduce_to(
           ds_reduce_util.ReduceOp.SUM, grads_and_vars)
@@ -529,9 +531,22 @@ def all_reduce_fn(distribution, grads_and_vars):
     # replica context.
     # TODO(b/150507409): Do not switch to a cross-replica context once the bug
     # is fixed.
-    if grads_and_vars:
-      return distribute_ctx.get_replica_context().merge_call(
-          all_reduce_fn, args=(grads_and_vars,))
+    if filtered_grads_and_vars:
+      reduced = distribute_ctx.get_replica_context().merge_call(
+          all_reduce_fn, args=(filtered_grads_and_vars,))
+    else:
+      reduced = []
+    # Copy 'reduced' but add None gradients back in
+    reduced_with_nones = []
+    reduced_pos = 0
+    for g, _ in grads_and_vars:
+      if g is None:
+        reduced_with_nones.append(None)
+      else:
+        reduced_with_nones.append(reduced[reduced_pos])
+        reduced_pos += 1
+    assert reduced_pos == len(reduced), "Failed to add all gradients"
+    return reduced_with_nones
 
   def _distributed_apply(self, distribution, grads_and_vars, name, apply_state):
     """`apply_gradients` using a `DistributionStrategy`."""
