with strategy.scope():
    # Create an mnist float model with the specified float state.
    model = FloatModel()
    optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)

  def train_step(features):
    inputs = tf.image.convert_image_dtype(
        features['image'], dtype=tf.float32, saturate=False)
    labels = tf.one_hot(features['label'], num_classes)

    with tf.GradientTape() as tape:

  //  bazel run -c opt --config=cuda ${TARGET_PATH}:resnet_imagenet_main       \
  //    -- --skip_eval --num_gpus=1 --dtype=fp16 --batch_size=192              \
  //    --train_steps=210 --enable_xla --enable_eager=true
  //
  // At CL 245846452
  TF_ASSERT_OK(RunAutoClusteringTestWithPbtxt("keras_imagenet_main"));
}

TEST_F(AutoClusteringTestImpl, KerasImagenetMainGraphMode) {

        deprecated and will be removed in a future release.
    *   Metrics update and collection logic in default `Model.train_step()` is
        now customizable via overriding `Model.compute_metrics()`.
    *   Losses computation logic in default `Model.train_step()` is now
        customizable via overriding `Model.compute_loss()`.
    *   `jit_compile` added to `Model.compile()` on an opt-in basis to compile