lmflow.pipeline.utils.peft_trainer
==================================

.. py:module:: lmflow.pipeline.utils.peft_trainer

.. autoapi-nested-parse::

   Trainer for Peft models

   ..
       !! processed by numpydoc !!


Classes
-------

.. autoapisummary::

   lmflow.pipeline.utils.peft_trainer.PeftTrainer
   lmflow.pipeline.utils.peft_trainer.PeftSavingCallback


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

.. py:class:: PeftTrainer

   Bases: :py:obj:`transformers.Trainer`


   .. py:method:: _save_checkpoint(_, trial, metrics=None)

      
      Don't save base model, optimizer etc.
      but create checkpoint folder (needed for saving adapter) 
















      ..
          !! processed by numpydoc !!


.. py:class:: PeftSavingCallback

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


   
   Correctly save PEFT model and not full model 
















   ..
       !! processed by numpydoc !!

   .. py:method:: _save(model, folder)


   .. py:method:: on_train_end(args: transformers.training_args.TrainingArguments, state: transformers.trainer_callback.TrainerState, control: transformers.trainer_callback.TrainerControl, **kwargs)

      
      Save final best model adapter 
















      ..
          !! processed by numpydoc !!


   .. py:method:: on_epoch_end(args: transformers.training_args.TrainingArguments, state: transformers.trainer_callback.TrainerState, control: transformers.trainer_callback.TrainerControl, **kwargs)

      
      Save intermediate model adapters in case of interrupted training 
















      ..
          !! processed by numpydoc !!


   .. py:method:: on_save(args: transformers.training_args.TrainingArguments, state: transformers.trainer_callback.TrainerState, control: transformers.trainer_callback.TrainerControl, **kwargs)