Internal change

PiperOrigin-RevId: 300447325

Author: Xhark

official/benchmark/keras_imagenet_benchmark.py

@@ -1305,7 +1305,6 @@ def __init__(self, root_data_dir=None, **kwargs):
         'model': 'resnet50_v1.5',
         'optimizer': 'mobilenet_default',
         'initial_learning_rate_per_sample': 0.0000039,
-        'use_tf_keras_layers': True,
         'pretrained_filepath': tf.train.latest_checkpoint(
             os.path.join(root_data_dir, 'resnet50')),
         'pruning_begin_step': 0,
@@ -1369,7 +1368,6 @@ def __init__(self, **kwargs):
     default_flags = {
         'model': 'resnet50_v1.5',
         'optimizer': 'mobilenet_default',
-        'use_tf_keras_layers': True,
     }
     super(Resnet50KerasPruningBenchmarkReal, self).__init__(
         default_flags=default_flags, **kwargs)

official/vision/image_classification/common.py

@@ -275,12 +275,6 @@ def define_keras_flags(
       help='Whether to build a tf.while_loop inside the training loop on the '
       'host. Setting it to True is critical to have peak performance on '
       'TPU.')
-  flags.DEFINE_boolean(
-      name='use_tf_keras_layers', default=False,
-      help='Whether to use tf.keras.layers instead of tf.python.keras.layers.'
-      'It only changes imagenet resnet model layers for now. This flag is '
-      'a temporal flag during transition to tf.keras.layers. Do not use this '
-      'flag for external usage. this will be removed shortly.')
 
   if model:
     flags.DEFINE_string('model', 'resnet50_v1.5',

official/vision/image_classification/resnet/resnet_model.py

@@ -31,7 +31,6 @@
 
 from tensorflow.python.keras import backend
 from tensorflow.python.keras import initializers
-from tensorflow.python.keras import layers as tf_python_keras_layers
 from tensorflow.python.keras import models
 from tensorflow.python.keras import regularizers
 from official.vision.image_classification import imagenet_preprocessing
@@ -40,30 +39,7 @@
 BATCH_NORM_DECAY = 0.9
 BATCH_NORM_EPSILON = 1e-5
 
-layers = tf_python_keras_layers
-
-
-def change_keras_layer(use_tf_keras_layers=False):
-  """Change layers to either tf.keras.layers or tf.python.keras.layers.
-
-  Layer version of  tf.keras.layers is depends on tensorflow version, but
-  tf.python.keras.layers checks environment variable TF2_BEHAVIOR.
-  This function is a temporal function to use tf.keras.layers.
-  Currently, tf v2 batchnorm layer is slower than tf v1 batchnorm layer.
-  this function is useful for tracking benchmark result for each version.
-  This function will be removed when we use tf.keras.layers as default.
-
-  TODO(b/146939027): Remove this function when tf v2 batchnorm reaches training
-  speed parity with tf v1 batchnorm.
-
-  Args:
-      use_tf_keras_layers: whether to use tf.keras.layers.
-  """
-  global layers
-  if use_tf_keras_layers:
-    layers = tf.keras.layers
-  else:
-    layers = tf_python_keras_layers
+layers = tf.keras.layers
 
 
 def _gen_l2_regularizer(use_l2_regularizer=True):

official/vision/image_classification/resnet_imagenet_main.py

@@ -183,7 +183,6 @@ def run(flags_obj):
       model = trivial_model.trivial_model(
           imagenet_preprocessing.NUM_CLASSES)
     elif flags_obj.model == 'resnet50_v1.5':
-      resnet_model.change_keras_layer(flags_obj.use_tf_keras_layers)
       model = resnet_model.resnet50(
           num_classes=imagenet_preprocessing.NUM_CLASSES)
     elif flags_obj.model == 'mobilenet':

official/vision/image_classification/resnet_imagenet_test.py

@@ -50,7 +50,6 @@ class KerasImagenetTest(tf.test.TestCase):
           "-model", "resnet50_v1.5",
           "-optimizer", "resnet50_default",
           "-pruning_method", "polynomial_decay",
-          "-use_tf_keras_layers", "true",
       ],
       "mobilenet": [
           "-model", "mobilenet",

official/vision/image_classification/resnet_runnable.py

@@ -70,7 +70,6 @@ def __init__(self, flags_obj, time_callback, epoch_steps):
     else:
       self.input_fn = imagenet_preprocessing.input_fn
 
-    resnet_model.change_keras_layer(flags_obj.use_tf_keras_layers)
     self.model = resnet_model.resnet50(
         num_classes=imagenet_preprocessing.NUM_CLASSES,
         batch_size=flags_obj.batch_size,