deepfakes/faceswap · error · ValueError

'{name}' is not a valid optimizer. Select from {list(_OPTIMI

Error message

'{name}' is not a valid optimizer. Select from {list(_OPTIMIZERS)}

What it means

Raised when building the training optimizer if config.optimizer() does not match any key in the _OPTIMIZERS registry in lib/training/optimizer.py. The optimizer name is read straight from the user's training configuration, so any unrecognised string aborts training at startup with this ValueError.

Source

Thrown at lib/training/optimizer.py:224

        return retval

    def _get_optimizer(self, model: K_Model, config: type[OptConfig]) -> torch.optim.Optimizer:
        """Obtain the configured optimizer the given configuration file options

        Parameters
        ----------
        model
            The keras model that is to be trained
        config
            The optimizer user configuration options

        Returns
        -------
        The requested configured optimizer
        """
        name = config.optimizer()
        if name not in _OPTIMIZERS:
            raise ValueError(f"'{name}' is not a valid optimizer. Select from {list(_OPTIMIZERS)}")
        optimizer = _OPTIMIZERS[name]

        retval = optimizer(self._get_parameter_groups(model, config.weight_decay()),
                           lr=config.learning_rate(),
                           **self._get_optimizer_kwargs(config))
        logger.debug("[Optimizer] Got optimizer '%s': %s", name, retval)
        return retval

    def _get_parameter_groups(self, model: K_Model, weight_decay: float
                              ) -> tuple[dict[T.Literal["params", "weight_decay"],
                                              list[nn.Parameter] | float],
                                         dict[T.Literal["params", "weight_decay"],
                                              list[nn.Parameter] | float]]:
        """Obtain the parameter groups from within the keras model

        Parameters
        ----------
        model

View on GitHub (pinned to f530cb7508)

Solutions

1. Set the optimizer in the training config to one of the names printed in the error (the keys of _OPTIMIZERS).
2. Pick the value via the GUI Train > Configure dialog so only valid names are offered.
3. Delete/regenerate the stale training config after upgrading Faceswap so it is rewritten with the currently valid options.

Example fix

# before
[optimizer.optimizer] = adamax

# after
[optimizer.optimizer] = adam

from lib.training.optimizer import _OPTIMIZERS  # module-level registry
assert config_optimizer_name in _OPTIMIZERS, f"pick from {list(_OPTIMIZERS)}"

def is_valid_optimizer(name: str, registry: dict) -> bool:
    return name in registry

When it happens

Trigger: Training is launched with an optimizer name in the training config that is not registered (typos like 'ADAM', 'adamw' when unsupported, or names removed after a Faceswap upgrade, e.g. the Keras-to-PyTorch backend migration).

Common situations: Hand-edited config files; configs carried across versions where the optimizer list changed; copying a config from a tutorial/forum listing optimizers this build does not include.

Related errors

• '{method}' is not a valid clipping method. Select from {list
• Penalized Mask Loss has been selected but you have not chose
• You have selected the mask type '{mask_type}' but at least o
• Load weights selected, but the path '{weights_file}' does no
• 'Learn Mask' has been selected but you have not chosen a Mas

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/067d5b7546f49820. Report an issue: GitHub.