lightning/docs/source-pytorch/upgrade/sections/1_4_advanced.rst

44 lines
1.5 KiB
ReStructuredText

.. list-table:: adv. user 1.4
:widths: 40 40 20
:header-rows: 1
* - If
- Then
- Ref
* - called ``ModelCheckpoint.save_function``
- now call ``Trainer.save_checkpoint``
- `PR7201`_
* - accessed the ``Trainer.running_sanity_check`` property
- now access the ``Trainer.sanity_checking`` property
- `PR4945`_
* - used ``LightningModule.grad_norm``
- now use the ``pl.utilities.grad_norm`` utility function instead
- `PR7292`_
* - used ``TrainerTrainingTricksMixin.detect_nan_tensors``
- now use ``pl.utilities.grads.grad_norm``
- `PR6834`_
* - used ``TrainerTrainingTricksMixin.print_nan_gradients``
- now use ``pl.utilities.finite_checks.print_nan_gradients``
- `PR6834`_
* - If you relied on ``TrainerLoggingMixin.metrics_to_scalars``
- now use ``pl.utilities.metrics.metrics_to_scalars``
- `PR7180`_
* - selected the i-th GPU with ``Trainer(gpus="i,j")``
- now this will set the number of GPUs, just like passing ``Trainer(devices=i)``, you can still select the specific GPU by setting the ``CUDA_VISIBLE_DEVICES=i,j`` environment variable
- `PR6388`_
.. _pr7201: https://github.com/Lightning-AI/lightning/pull/7201
.. _pr4945: https://github.com/Lightning-AI/lightning/pull/4945
.. _pr7292: https://github.com/Lightning-AI/lightning/pull/7292
.. _pr6834: https://github.com/Lightning-AI/lightning/pull/6834
.. _pr7180: https://github.com/Lightning-AI/lightning/pull/7180
.. _pr6388: https://github.com/Lightning-AI/lightning/pull/6388