microsoft/VibeVoice · error · ValueError

Cannot use `timesteps` with `config.use_karras_sigmas = True

Error message

Cannot use `timesteps` with `config.use_karras_sigmas = True`

What it means

Custom `timesteps` and the Karras sigma schedule (`use_karras_sigmas=True` in the scheduler config) are mutually exclusive: Karras sigmas are generated from the trained sigma schedule and then mapped back to timesteps, so a user-supplied timestep list would be overwritten/inconsistent. `set_timesteps` rejects the combination with this ValueError.

Source

Thrown at vibevoice/schedule/dpm_solver.py:345

        """
        Sets the discrete timesteps used for the diffusion chain (to be run before inference).

        Args:
            num_inference_steps (`int`):
                The number of diffusion steps used when generating samples with a pre-trained model.
            device (`str` or `torch.device`, *optional*):
                The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
            timesteps (`List[int]`, *optional*):
                Custom timesteps used to support arbitrary timesteps schedule. If `None`, timesteps will be generated
                based on the `timestep_spacing` attribute. If `timesteps` is passed, `num_inference_steps` and `sigmas`
                must be `None`, and `timestep_spacing` attribute will be ignored.
        """
        if num_inference_steps is None and timesteps is None:
            raise ValueError("Must pass exactly one of `num_inference_steps` or `timesteps`.")
        if num_inference_steps is not None and timesteps is not None:
            raise ValueError("Can only pass one of `num_inference_steps` or `custom_timesteps`.")
        if timesteps is not None and self.config.use_karras_sigmas:
            raise ValueError("Cannot use `timesteps` with `config.use_karras_sigmas = True`")
        if timesteps is not None and self.config.use_lu_lambdas:
            raise ValueError("Cannot use `timesteps` with `config.use_lu_lambdas = True`")

        if timesteps is not None:
            timesteps = np.array(timesteps).astype(np.int64)
        else:
            # Clipping the minimum of all lambda(t) for numerical stability.
            # This is critical for cosine (squaredcos_cap_v2) noise schedule.
            clipped_idx = torch.searchsorted(torch.flip(self.lambda_t, [0]), self.config.lambda_min_clipped)
            last_timestep = ((self.config.num_train_timesteps - clipped_idx).numpy()).item()

            # "linspace", "leading", "trailing" corresponds to annotation of Table 2. of https://arxiv.org/abs/2305.08891
            if self.config.timestep_spacing == "linspace":
                timesteps = (
                    np.linspace(0, last_timestep - 1, num_inference_steps + 1)
                    .round()[::-1][:-1]
                    .copy()
                    .astype(np.int64)

View on GitHub (pinned to 94da20d98b)

Solutions

1. Construct the scheduler with `use_karras_sigmas=False` if you need custom timesteps.
2. Or keep Karras sigmas and pass `num_inference_steps` instead of `timesteps`.
3. If you need both (custom grid + Karras spacing), compute the Karras sigmas yourself and set `scheduler.sigmas`/`timesteps` manually after set_timesteps.

Example fix

# before
sched = DPMSolverMultistepScheduler(..., use_karras_sigmas=True)
sched.set_timesteps(timesteps=[900, 700, 400, 100])

# after
sched = DPMSolverMultistepScheduler(..., use_karras_sigmas=False)
sched.set_timesteps(timesteps=[900, 700, 400, 100])

if timesteps is not None and scheduler.config.use_karras_sigmas:
    raise ValueError("Custom timesteps require a scheduler built with use_karras_sigmas=False")
scheduler.set_timesteps(num_inference_steps=n, timesteps=timesteps)

def can_use_custom_timesteps(scheduler) -> bool:
    return not (scheduler.config.use_karras_sigmas or scheduler.config.use_lu_lambdas)

When it happens

Trigger: `DPMSolverMultistepScheduler(..., use_karras_sigmas=True)` later followed by `scheduler.set_timesteps(timesteps=[...])`.

Common situations: SDEdit/img2img flows that specify timesteps while the scheduler was built with karras sigmas for quality; reusing one scheduler config across pipelines with different calling conventions.

Related errors

• Cannot use `timesteps` with `config.use_lu_lambdas = True`
• {self.config.timestep_spacing} is not supported. Please make
• Unsupported alpha_transform_type: {alpha_transform_type}
• {beta_schedule} is not implemented for {self.__class__}
• {algorithm_type} is not implemented for {self.__class__}

AI-assisted analysis of microsoft/VibeVoice@94da20d98b (2026-08-15). Data as JSON: /api/errors/6b8b046a95273c1d. Report an issue: GitHub.