sgl-project/sglang · error · ValueError

Unknown time_shift_type: {self.config.time_shift_type}

Error message

Unknown time_shift_type: {self.config.time_shift_type}

What it means

time_shift() dispatches on config.time_shift_type and raises for any value other than the two supported branches. It runs during set_timesteps, so a bad value can slip past the constructor if the config was mutated after construction.

Source

Thrown at python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_flow_match_euler_discrete.py:249

        while len(sigma.shape) < len(sample.shape):
            sigma = sigma.unsqueeze(-1)

        sample = sigma * noise + (1.0 - sigma) * sample

        return sample

    def _sigma_to_t(self, sigma: float) -> float:
        return sigma * self.config.num_train_timesteps

    def time_shift(
        self, mu: float, sigma: float, t: torch.Tensor | np.ndarray
    ) -> torch.Tensor | np.ndarray:
        if self.config.time_shift_type == "exponential":
            return self._time_shift_exponential(mu, sigma, t)
        elif self.config.time_shift_type == "linear":
            return self._time_shift_linear(mu, sigma, t)
        else:
            raise ValueError(f"Unknown time_shift_type: {self.config.time_shift_type}")

    def stretch_shift_to_terminal(self, t: torch.Tensor) -> torch.Tensor:
        r"""
        Stretches and shifts the timestep schedule to ensure it terminates at the configured `shift_terminal` config
        value.

        Reference:
        https://github.com/Lightricks/LTX-Video/blob/a01a171f8fe3d99dce2728d60a73fecf4d4238ae/ltx_video/schedulers/rf.py#L51

        Args:
            t (`torch.Tensor`):
                A tensor of timesteps to be stretched and shifted.

        Returns:
            `torch.Tensor`:
                A tensor of adjusted timesteps such that the final value equals `self.config.shift_terminal`.
        """
        one_minus_z = 1 - t

View on GitHub (pinned to 0132848349)

Solutions

1. Reset time_shift_type to 'exponential' or 'linear' before set_timesteps
2. Re-instantiate the scheduler with a valid type so constructor validation runs
3. Avoid mutating scheduler.config after construction

Example fix

# before
scheduler.config.time_shift_type = "sigmoid"  # later set_timesteps -> error
# after
scheduler.config.time_shift_type = "linear"

assert scheduler.config.time_shift_type in {"exponential", "linear"}
scheduler.set_timesteps(steps, mu=mu)

When it happens

Trigger: Constructing the scheduler, then setting scheduler.config.time_shift_type = 'custom' before set_timesteps; bypassing the constructor check by mutating a loaded config dict.

Common situations: Runtime config overrides, patches applied by pipelines, or config deserialization paths that skip __init__ validation.

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• `time_shift_type` must either be 'exponential' or 'linear'.
• {response.error}
• action policy returned no output
• Could not connect to remote scheduler at {self.server_args.s
• {output_batch.error}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/2627dba252f3f808. Report an issue: GitHub.