lllyasviel/ControlNet · error · NotImplementedError

Parameterization {self.parameterization} not yet supported

Error message

Parameterization {self.parameterization} not yet supported

What it means

p_losses must know what the model predicts: 'eps' (noise), 'x0' (clean data), or 'v' (velocity parameterization). An unrecognized parameterization string in the config raises NotImplementedError before loss computation.

Source

Thrown at ldm/models/diffusion/ddpm.py:395

        else:
            raise NotImplementedError("unknown loss type '{loss_type}'")

        return loss

    def p_losses(self, x_start, t, noise=None):
        noise = default(noise, lambda: torch.randn_like(x_start))
        x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
        model_out = self.model(x_noisy, t)

        loss_dict = {}
        if self.parameterization == "eps":
            target = noise
        elif self.parameterization == "x0":
            target = x_start
        elif self.parameterization == "v":
            target = self.get_v(x_start, noise, t)
        else:
            raise NotImplementedError(f"Parameterization {self.parameterization} not yet supported")

        loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3])

        log_prefix = 'train' if self.training else 'val'

        loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()})
        loss_simple = loss.mean() * self.l_simple_weight

        loss_vlb = (self.lvlb_weights[t] * loss).mean()
        loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb})

        loss = loss_simple + self.original_elbo_weight * loss_vlb

        loss_dict.update({f'{log_prefix}/loss': loss})

        return loss, loss_dict

    def forward(self, x, *args, **kwargs):

View on GitHub (pinned to ed85cd1e25)

Solutions

1. Set parameterization: eps (standard), x0, or v in the diffusion config
2. If the key is missing in an old config, add it explicitly rather than relying on defaults
3. For a new parameterization, extend p_losses with your target computation

Example fix

# before
parameterization: epsilon
# after
parameterization: eps

assert parameterization in ('eps', 'x0', 'v'), f"unsupported parameterization {parameterization!r}"

def is_valid_parameterization(p: str) -> bool:
    return p in ('eps', 'x0', 'v')

When it happens

Trigger: Running training/inference with parameters.parameterization set to anything other than 'eps', 'x0', or 'v' (e.g. 'epsilon', 'v-pred', or None).

Common situations: Copying configs from v-prediction forks with different naming; hand-merging YAML configs; older configs omitting parameterization while code requires an exact known string.

Related errors

• unknown loss type '{loss_type}'
• Unsupported noise schedule {}. The schedule needs to be 'dis
• resize_method {self.__resize_method} not implemented
• provide num_res_blocks either as an int (globally constant)
• encoder_posterior of type '{type(encoder_posterior)}' not ye

AI-assisted analysis of lllyasviel/ControlNet@ed85cd1e25 (2026-08-27). Data as JSON: /api/errors/5d31d9b63ba5a99d. Report an issue: GitHub.