ut3_load
========

.. py:function:: t3toolbox.backend.ut3_constructors.ut3_load(file, use_jax = False)

   .. code-block:: python

      def ut3_load(
              file,                  # path or open file object to read the .npz from
              use_jax: bool = False, # chooses the SUPERCORE type; masks always come back numpy (host) bool
      ) -> UT3Data:


   Load a uniform Tucker tensor train from a ``.npz`` file written by :py:func:`ut3_save`.

   The supercores follow ``use_jax``; the masks stay **numpy (host) bool** regardless -- a jax mask is a
   tracer under jit and breaks the layer (``docs/contributor/uniform_pytree_composition.md``). ``np.load`` returns
   the masks with their saved bool dtype; we only convert the supercores.