Back to Pytorch Lightning

1 4 Advanced

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

2.6.41.6 KB
Original Source

.. 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/pytorch-lightning/pull/7201 .. _pr4945: https://github.com/Lightning-AI/pytorch-lightning/pull/4945 .. _pr7292: https://github.com/Lightning-AI/pytorch-lightning/pull/7292 .. _pr6834: https://github.com/Lightning-AI/pytorch-lightning/pull/6834 .. _pr7180: https://github.com/Lightning-AI/pytorch-lightning/pull/7180 .. _pr6388: https://github.com/Lightning-AI/pytorch-lightning/pull/6388