def from_instances(cls, model: Tacotron2, optimizer: Adam,
hparams: HParams, iteration: int, symbols: SymbolIdDict,
accents: AccentsDict, speakers: SpeakersDict):
result = cls(state_dict=model.state_dict(),
optimizer=optimizer.state_dict(),
learning_rate=hparams.learning_rate,
iteration=iteration,
hparams=asdict(hparams),
symbols=symbols.raw(),
accents=accents.raw(),
speakers=speakers.raw())
return result
def convert_v1_to_v2_model(old_model_path: str,
custom_hparams: Optional[Dict[str, str]],
speakers: SpeakersDict, accents: AccentsDict,
symbols: SymbolIdDict):
checkpoint_dict = torch.load(old_model_path, map_location='cpu')
hparams = HParams(n_speakers=len(speakers),
n_accents=len(accents),
n_symbols=len(symbols))
hparams = overwrite_custom_hparams(hparams, custom_hparams)
chp = CheckpointTacotron(state_dict=checkpoint_dict["state_dict"],
optimizer=checkpoint_dict["optimizer"],
learning_rate=checkpoint_dict["learning_rate"],
iteration=checkpoint_dict["iteration"] + 1,
hparams=asdict(hparams),
speakers=speakers.raw(),
symbols=symbols.raw(),
accents=accents.raw())
new_model_path = f"{old_model_path}_{get_pytorch_filename(chp.iteration)}"
chp.save(new_model_path, logging.getLogger())