ds_train = tfds.load('mnist', split='train', shuffle_files=True)
  ds_train = ds_train.shuffle(1024).batch(batch_size).prefetch(64)
  ds_train = strategy.experimental_distribute_dataset(ds_train)

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

        causes of issues involving infinities and `NaN`s.
*   `tf.distribute`
    *   Custom training loop support on TPUs and TPU pods is available through
        `strategy.experimental_distribute_dataset`,
        `strategy.experimental_distribute_datasets_from_function`,
        `strategy.experimental_run_v2`, `strategy.reduce`.
    *   Support for a global distribution strategy through