sgl-project/sglang · error · ValueError

`sigmas` and `timesteps` should have the same length

Error message

`sigmas` and `timesteps` should have the same length

What it means

When both sigmas and timesteps are provided to set_timesteps they must describe the same schedule step-for-step; mismatched lengths raise this validation error before anything is assigned.

Source

Thrown at python/sglang/multimodal_gen/runtime/models/schedulers/scheduling_flow_match_euler_discrete.py:311

            mu (`float`, *optional*):
                Determines the amount of shifting applied to sigmas when performing resolution-dependent timestep
                shifting.
            timesteps (`List[float]`, *optional*):
                Custom values for timesteps to be used for each diffusion step. If `None`, the timesteps are computed
                automatically.
        """

        if self.config.use_dynamic_shifting and mu is None:
            raise ValueError(
                "`mu` must be passed when `use_dynamic_shifting` is set to be `True`"
            )

        if (
            sigmas is not None
            and timesteps is not None
            and len(sigmas) != len(timesteps)
        ):
            raise ValueError("`sigmas` and `timesteps` should have the same length")

        if num_inference_steps is not None:
            if (sigmas is not None and len(sigmas) != num_inference_steps) or (
                timesteps is not None and len(timesteps) != num_inference_steps
            ):
                raise ValueError(
                    "`sigmas` and `timesteps` should have the same length as num_inference_steps, if `num_inference_steps` is provided"
                )
        else:
            if sigmas is not None:
                num_inference_steps = len(sigmas)
            elif timesteps is not None:
                num_inference_steps = len(timesteps)
            else:
                raise ValueError(
                    "Either num_inference_steps, sigmas, or timesteps must be provided"
                )

View on GitHub (pinned to 0132848349)

Solutions

1. Make len(sigmas) == len(timesteps); trim or pad the terminal sigma accordingly
2. Generate both from the same num_inference_steps and pass only one of them
3. Validate lengths in your schedule-generation code before calling

Example fix

# before
scheduler.set_timesteps(50, sigmas=sigmas, timesteps=ts)  # len 51 vs 50
# after
scheduler.set_timesteps(50, sigmas=sigmas[:-1], timesteps=ts)

assert sigmas is None or timesteps is None or len(sigmas) == len(timesteps)

When it happens

Trigger: Passing 50 sigmas (including terminal) with 49 timesteps, or reusing sigmas from one step count with timesteps from another.

Common situations: Custom schedules where users forget the terminal sigma (length N+1) versus N timesteps; mixing schedules generated for different num_inference_steps.

Related errors

• `sigmas` and `timesteps` should have the same length as num_
• kv-canary: {name} length must be {expected}, got {actual}
• pairs must be a torch.Tensor of shape [N, 2]
• pairs must be a torch.Tensor
• pairs length must be greater than 0

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/2453a2bceb14771d. Report an issue: GitHub.