`(batch_size, 1)`. This enables `Model` subclasses to process scalar
        data in their `train_step()`/`test_step()`/`predict_step()` methods. \
        Note that this change may break certain subclassed models. You can
        revert back to the previous behavior by adding upranking yourself in the
        `train_step()`/`test_step()`/`predict_step()` methods, e.g. `if
        x.shape.rank == 1: x = tf.expand_dims(x, axis=-1)`. Functional models as

2F3A          ; mapped                 ; 5F61          # 3.0  KANGXI RADICAL BRISTLE
2F3B          ; mapped                 ; 5F73          # 3.0  KANGXI RADICAL STEP
2F3C          ; mapped                 ; 5FC3          # 3.0  KANGXI RADICAL HEART
2F3D          ; mapped                 ; 6208          # 3.0  KANGXI RADICAL HALBERD