| from pytorch_lightning import Trainer |
| from pytorch_lightning.callbacks import ModelCheckpoint, LearningRateMonitor |
| from pytorch_lightning.loggers import TensorBoardLogger |
|
|
| from s1.data_module import QuantizeDataModule |
| from s1.trainer import Wav2TTS |
| from pytorch_lightning.strategies import DDPStrategy |
| import json |
|
|
|
|
| class AttrDict(dict): |
| def __init__(self, *args, **kwargs): |
| super(AttrDict, self).__init__(*args, **kwargs) |
| self.__dict__ = self |
|
|
| args = AttrDict() |
|
|
| arg_dict = json.load(open('configs/mq.json', 'r')) |
| args.__dict__.update(arg_dict) |
|
|
| fname_prefix = f'' |
|
|
| if args.accelerator == 'ddp': |
| args.accelerator = DDPStrategy(find_unused_parameters=True) |
|
|
| checkpoint_callback = ModelCheckpoint( |
| dirpath=args.saving_path, |
| filename=(fname_prefix+'{epoch}-{step}'), |
| every_n_train_steps=(None if args.val_check_interval == 1.0 else args.val_check_interval), |
| every_n_epochs=(None if args.check_val_every_n_epoch == 1 else args.check_val_every_n_epoch), |
| verbose=True, |
| save_last=True |
| ) |
|
|
| logger = TensorBoardLogger(args.sampledir, name="VQ-TTS", version=args.version) |
| lr_monitor = LearningRateMonitor(logging_interval='step') |
|
|
| wrapper = Trainer( |
| precision=args.precision, |
| callbacks=[checkpoint_callback, lr_monitor], |
| val_check_interval=args.val_check_interval, |
| num_sanity_val_steps=0, |
| max_steps=args.training_step, |
| devices=(-1 if args.distributed else 1), |
| strategy=(args.accelerator if args.distributed else None), |
| use_distributed_sampler=False, |
| accumulate_grad_batches=args.accumulate_grad_batches, |
| logger=logger, |
| check_val_every_n_epoch=None |
| ) |
|
|
| training_wrapper = Wav2TTS(args) |
| |
| |
| data_module = QuantizeDataModule(args) |
| wrapper.fit(training_wrapper, datamodule=data_module, ckpt_path=args.resume_checkpoint) |
