microsoft/VibeVoice · error · ValueError

Unsupported alpha_transform_type: {alpha_transform_type}

Error message

Unsupported alpha_transform_type: {alpha_transform_type}

What it means

`betas_for_alpha_bar()` is a helper that builds a noise schedule from a parametric alpha_bar function; this copy supports only `cosine`, `exp`, `cauchy`, and `laplace` transform types (dpm_solver.py:51-73). An unknown string raises ValueError immediately. Upstream diffusers supports the same set plus historically `bspline`, so code ported from other schedulers can pass an unsupported name.

Source

Thrown at vibevoice/schedule/dpm_solver.py:76

        def alpha_bar_fn(t):
            return math.exp(t * -12.0)

    elif alpha_transform_type == "cauchy":
        # µ + γ tan (π (0.5 - x))  γ = 1, µ = 3
        # alpha^2 = 1-1/(exp(λ)+1)
        def alpha_bar_fn(t, gamma=1, mu=3):
            snr = mu + gamma * math.tan(math.pi * (0.5 - t) * 0.9)
            return 1 - 1 / (math.exp(snr) + 1.1)

    elif alpha_transform_type == "laplace":
        # µ − bsgn(0.5 − t) log(1 − 2|t − 0.5|) µ = 0, b = 1
        def alpha_bar_fn(t, mu=0, b=1):
            snr = mu - b * math.copysign(1, 0.5 - t) * math.log(1 - 2 * abs(t - 0.5) * 0.98)
            return 1 - 1 / (math.exp(snr) + 1.02)

    else:
        raise ValueError(f"Unsupported alpha_transform_type: {alpha_transform_type}")

    betas = []
    for i in range(num_diffusion_timesteps):
        t1 = i / num_diffusion_timesteps
        t2 = (i + 1) / num_diffusion_timesteps
        betas.append(min(1 - alpha_bar_fn(t2) / alpha_bar_fn(t1), max_beta))
    return torch.tensor(betas, dtype=torch.float32)


# Copied from diffusers.schedulers.scheduling_ddim.rescale_zero_terminal_snr
def rescale_zero_terminal_snr(betas):
    """
    Rescales betas to have zero terminal SNR Based on https://arxiv.org/pdf/2305.08891.pdf (Algorithm 1)


    Args:
        betas (`torch.Tensor`):
            the betas that the scheduler is being initialized with.

View on GitHub (pinned to 94da20d98b)

Solutions

1. Use one of the supported values: "cosine", "exp", "cauchy", or "laplace" (all lowercase).
2. If you ported the call from diffusers expecting `bspline`, reimplement it locally: define your own alpha_bar_fn and compute betas with the same min(1 - fn(t2)/fn(t1), max_beta) loop.
3. Check for typos/case in scheduler config values loaded from YAML/JSON.

Example fix

# before
betas = betas_for_alpha_bar(1000, alpha_transform_type="bspline")

# after
betas = betas_for_alpha_bar(1000, alpha_transform_type="cosine")

SUPPORTED = {"cosine", "exp", "cauchy", "laplace"}
assert alpha_transform_type in SUPPORTED, (
    f"alpha_transform_type must be one of {sorted(SUPPORTED)}, got {alpha_transform_type!r}"
)
betas = betas_for_alpha_bar(1000, alpha_transform_type=alpha_transform_type)

def is_valid_alpha_transform(v) -> bool:
    return isinstance(v, str) and v in {"cosine", "exp", "cauchy", "laplace"}

When it happens

Trigger: Calling `betas_for_alpha_bar(N, alpha_transform_type="bspline")` or any string outside {cosine, exp, cauchy, laplace}; indirectly via a scheduler constructor only if beta_schedule maps there (the scheduler itself routes cosine/cauchy/laplace, so this is almost always a direct helper call).

Common situations: Porting schedule code from other diffusion repos (some use `bspline` or custom names), typos like `cosine2`/`Cosine` (case-sensitive), or copying config YAML from a model trained with a different scheduler family.

Related errors

• {beta_schedule} is not implemented for {self.__class__}
• {algorithm_type} is not implemented for {self.__class__}
• {solver_type} is not implemented for {self.__class__}
• `final_sigmas_type` {final_sigmas_type} is not supported for
• Cannot use `timesteps` with `config.use_karras_sigmas = True

AI-assisted analysis of microsoft/VibeVoice@94da20d98b (2026-08-15). Data as JSON: /api/errors/9f35511f3191808e. Report an issue: GitHub.