how to convert custom ckpt to bin?

[SaltedSlark]

Hi @hubertsiuzdak, I am trying to figure out how to convert my ckpt to pytorch_model.bin so that I can load model by vocos.pretrained or any idea to load ckpt for inferencing directly?

[taalua]

Hey,

You can do this way:
import torch
ckpt = torch.load('last.ckpt')
torch.save(ckpt['state_dict'], 'pytorch.bin')

then when loading the model:

    def from_model(cls, model_path: str, config_path: str) -> Vocos:
        """
        Class method to create a new Vocos saved model
        """
        model = cls.from_hparams(config_path)
        state_dict = torch.load(model_path, map_location="cpu")
        if isinstance(model.feature_extractor, EncodecFeatures):
            encodec_parameters = {
                "feature_extractor.encodec." + key: value
                for key, value in model.feature_extractor.encodec.state_dict().items()
            }
            state_dict.update(encodec_parameters)
        model.load_state_dict(state_dict, strict=False)
        model.eval()
        return model

[SaltedSlark]

Hey,

You can do this way: import torch ckpt = torch.load('last.ckpt') torch.save(ckpt['state_dict'], 'pytorch.bin')

then when loading the model:

    def from_model(cls, model_path: str, config_path: str) -> Vocos:
        """
        Class method to create a new Vocos saved model
        """
        model = cls.from_hparams(config_path)
        state_dict = torch.load(model_path, map_location="cpu")
        if isinstance(model.feature_extractor, EncodecFeatures):
            encodec_parameters = {
                "feature_extractor.encodec." + key: value
                for key, value in model.feature_extractor.encodec.state_dict().items()
            }
            state_dict.update(encodec_parameters)
        model.load_state_dict(state_dict, strict=False)
        model.eval()
        return model

Thanks for ur reply! I'll have a try 👍

[dgo2dance]

+1