used to shard the checkpoint with a maximum shard file size. Users with
      advanced use cases can also write their own custom
      `tf.train.experimental.ShardingCallback`s.

* `tf.train.CheckpointOptions`
    * Added `experimental_skip_slot_variables` (a boolean option) to skip
    restoring of optimizer slot variables in a checkpoint.

*   `tf.saved_model.SaveOptions`