lmflow.pipeline.utils.lisa_trainer
==================================

.. py:module:: lmflow.pipeline.utils.lisa_trainer


Classes
-------

.. autoapisummary::

   lmflow.pipeline.utils.lisa_trainer.DynamicLayerActivationCallback


Module Contents
---------------

.. py:class:: DynamicLayerActivationCallback(n_layers: int, interval_steps: int, model: transformers.PreTrainedModel, lisa_layers_attribute: Optional[str] = None)

   Bases: :py:obj:`transformers.trainer_callback.TrainerCallback`


   .. py:attribute:: n_layers


   .. py:attribute:: interval_steps


   .. py:attribute:: model


   .. py:attribute:: total_layers


   .. py:attribute:: active_layers_indices
      :value: []



   .. py:method:: freeze_all_layers()


   .. py:method:: on_step_begin(args, state, control, **kwargs)


   .. py:method:: switch_active_layers()