Return typeNoneon_epoch_endModelHooks.on_epoch_end()Called when either of train/val/test epoch ends.Return typeNone74Chapter 7. LightningModule
PyTorch Lightning Documentation, Release 1.3.8on_train_epoch_startModelHooks.on_train_epoch_start()Called in the training loop at the very beginning of the epoch.Return typeNoneon_train_epoch_endModelHooks.on_train_epoch_end(unused=None)Called in the training loop at the very end of the epoch.To access all batch outputs at the end of the epoch, either:1. Implementtraining_epoch_endin the LightningModule OR2. Cache data across steps on the attribute(s) of theLightningModuleand access them in this hookon_validation_batch_startModelHooks.on_validation_batch_start(batch,batch_idx,dataloader_idx)Called in the validation loop before anything happens for that batch.Parameters•batch¶(Any) – The batched data as it is returned by the validation DataLoader.•batch_idx¶(int) – the index of the batch•dataloader_idx¶(int) – the index of the dataloaderReturn typeNoneon_validation_batch_endModelHooks.on_validation_batch_end(outputs,batch,batch_idx,dataloader_idx)Called in the validation loop after the batch.Parameters•outputs¶(Union[Tensor,Dict[str,Any],None]) – The outputs of valida-tion_step_end(validation_step(x))•batch¶(Any) – The batched data as it is returned by the validation DataLoader.•batch_idx¶(int) – the index of the batch•dataloader_idx¶(int) – the index of the dataloaderReturn typeNone7.4. LightningModule API75
PyTorch Lightning Documentation, Release 1.3.8on_validation_epoch_startModelHooks.on_validation_epoch_start()Called in the validation loop at the very beginning of the epoch.Return typeNoneon_validation_epoch_endModelHooks.on_validation_epoch_end()Called in the validation loop at the very end of the epoch.Return typeNoneon_post_move_to_deviceModelHooks.on_post_move_to_device()Called in theparameter_validationdecorator afterto()is called. This is a good place to tie weightsbetween modules after moving them to a device. Can be used when training models with weight sharing prop-erties on TPU.AddressesthehandlingofsharedweightsonTPU:TROUBLESHOOTING.md#xla-tensor-quirksExample:defon_post_move_to_device(self):self.decoder.weight=self.encoder.weightReturn typeNoneon_validation_model_evalModelHooks.on_validation_model_eval()Sets the model to eval during the val loopReturn typeNoneon_validation_model_trainModelHooks.on_validation_model_train()Sets the model to train during the val loopReturn typeNone76Chapter 7. LightningModule
PyTorch Lightning Documentation, Release 1.3.8on_test_model_evalModelHooks.on_test_model_eval()Sets the model to eval during the test loopReturn typeNoneon_test_model_trainModelHooks.on_test_model_train()Sets the model to train during the test loopReturn typeNoneoptimizer_stepLightningModule.optimizer_step(epoch=None,batch_idx=None,optimizer=None,opti-mizer_idx=None,optimizer_closure=None,on_tpu=None,using_native_amp=None,using_lbfgs=None)Override this method to adjust the default way theTrainercalls each optimizer. By default, Lightning callsstep()andzero_grad()as shown in the example once per optimizer.
Upload your study docs or become a
Course Hero member to access this document
Upload your study docs or become a
Course Hero member to access this document
End of preview. Want to read all 618 pages?
Upload your study docs or become a
Course Hero member to access this document
