def main(cfg): trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = HifiGanModel(cfg=cfg.model, trainer=trainer) epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(plugins=[DDPPlugin(find_unused_parameters=True)], **cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = MelGanModel(cfg=cfg.model, trainer=trainer) epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get('exp_manager', None)) model = TalkNetDursModel(cfg=cfg.model, trainer=trainer) lr_logger = pl.callbacks.LearningRateLogger() epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([lr_logger, epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = FastPitchModel(cfg=cfg.model, trainer=trainer) lr_logger = pl.callbacks.LearningRateMonitor() epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([lr_logger, epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get('exp_manager', None)) model = TalkNetPitchModel(cfg=cfg.model) trainer.callbacks.extend( [pl.callbacks.LearningRateMonitor(), LogEpochTimeCallback()]) # noqa trainer.fit(model)
def main(cfg): preprocess_linear_specs_dataset(**cfg.model.preprocessor) trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = EDMel2SpecModel(cfg=cfg.model, trainer=trainer) epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(plugins=[DDPPlugin(find_unused_parameters=True)], **cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = FastPitchHifiGanE2EModel(cfg=cfg.model, trainer=trainer) lr_logger = pl.callbacks.LearningRateMonitor() epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([lr_logger, epoch_time_logger]) trainer.fit(model)
def main(cfg): # Define the Lightning trainer trainer = pl.Trainer(**cfg.trainer) # exp_manager is a NeMo construct that helps with logging and checkpointing exp_manager(trainer, cfg.get("exp_manager", None)) # Define the Tacotron 2 model, this will construct the model as well as # define the training and validation dataloaders model = Tacotron2Model(cfg=cfg.model, trainer=trainer) # Let's add a few more callbacks lr_logger = pl.callbacks.LearningRateMonitor() epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([lr_logger, epoch_time_logger]) # Call lightning trainer's fit() to train the model trainer.fit(model)
def main(cfg): if hasattr(cfg.model.optim, 'sched'): logging.warning( "You are using an optimizer scheduler while finetuning. Are you sure this is intended?" ) if cfg.model.optim.lr > 1e-3 or cfg.model.optim.lr < 1e-5: logging.warning("The recommended learning rate for finetuning is 2e-4") trainer = pl.Trainer(**cfg.trainer) exp_manager(trainer, cfg.get("exp_manager", None)) model = FastPitchModel(cfg=cfg.model, trainer=trainer) model.maybe_init_from_pretrained_checkpoint(cfg=cfg) lr_logger = pl.callbacks.LearningRateMonitor() epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([lr_logger, epoch_time_logger]) trainer.fit(model)
def main(cfg): trainer = pl.Trainer(**cfg.trainer) epoch_time_logger = LogEpochTimeCallback() trainer.callbacks.extend([epoch_time_logger]) exp_manager(trainer, cfg.get("exp_manager", None)) if cfg.resume_from_ckpt is None: logging.info("Training UniGlow from scratch") model = UniGlowModel(cfg=cfg.model, trainer=trainer) else: logging.info("Fine-tuning UniGlow from {cfg.resume_from_ckpt}") model = UniGlowModel.restore_from(cfg.resume_from_ckpt) model.setup_training_data(cfg.model.train_ds) model.setup_validation_data(cfg.model.validation_ds) trainer.fit(model)