1 4 Regular - Pytorch Lightning

.. list-table:: reg. user 1.4 :widths: 40 40 20 :header-rows: 1

- If
- Then
- Ref
- relied on the outputs in your LightningModule.on_train_epoch_end or Callback.on_train_epoch_end hooks
- rely on either on_train_epoch_end or set outputs as attributes in your LightningModule instances and access them from the hook
- PR7339_
- accessed Trainer.truncated_bptt_steps
- switch to manual optimization
- PR7323_
- called LightningModule.write_predictions and LightningModule.write_predictions_dict
- rely on predict_step and Trainer.predict + callbacks to write out predictions
- PR7066_
- passed the period argument to the ModelCheckpoint callback
- pass the every_n_epochs argument to the ModelCheckpoint callback
- PR6146_
- passed the output_filename argument to Profiler
- now pass dirpath and filename, that is Profiler(dirpath=...., filename=...)
- PR6621_
- passed the profiled_functions argument in PytorchProfiler
- now pass the record_functions argument
- PR6349_
- relied on the @auto_move_data decorator to use the LightningModule outside of the Trainer for inference
- use Trainer.predict
- PR6993_
- implemented on_load_checkpoint with a checkpoint only argument, as in Callback.on_load_checkpoint(checkpoint)
- now update the signature to include pl_module and trainer, as in Callback.on_load_checkpoint(trainer, pl_module, checkpoint)
- PR7253_
- relied on pl.metrics
- now import separate package torchmetrics
- torchmetrics_
- accessed datamodule attribute of LightningModule, that is model.datamodule
- now access Trainer.datamodule, that is model.trainer.datamodule
- PR7168_