* @throws NullPointerException if any of {@code elements} is null
   * @since 7.0 (source-compatible since 2.0)
   */
  // This the best we could do to get copyOfEnumSet to compile in the mainline.
  // The suppression also covers the cast to E[], discussed below.
  // In the backport, we don't have those cases and thus don't need this suppression.
  // We keep it to minimize diffs.
  @SuppressWarnings("unchecked")

        the already-constructed model instead.
    *   Code that requires very tricky shape manipulation via converted op
        layers in order to work, where the Keras symbolic shape inference proves
        insufficient.
    *   Code that tries manually walking a `tf.keras.Model` layer by layer and
        assumes layers only ever have one positional argument. This assumption