sgl-project/sglang · error · ValueError

Passing integer indices as timesteps is not supported. Pass

Error message

Passing integer indices as timesteps is not supported. Pass one of `scheduler.timesteps` as a timestep.

What it means

step() rejects int/IntTensor/LongTensor timesteps because the flow-match scheduler treats timesteps as continuous float values (indexed lookup via _init_step_index would be ambiguous). Timesteps must be float values taken from scheduler.timesteps.

Source

Thrown at python/sglang/multimodal_gen/runtime/models/schedulers/hunyuan3d_scheduler.py:194

            self._step_index = self.index_for_timestep(timestep)
        else:
            self._step_index = self._begin_index

    def step(
        self,
        model_output: torch.FloatTensor,
        timestep: Union[float, torch.FloatTensor],
        sample: torch.FloatTensor,
        s_churn: float = 0.0,
        s_tmin: float = 0.0,
        s_tmax: float = float("inf"),
        s_noise: float = 1.0,
        generator: Optional[torch.Generator] = None,
        return_dict: bool = True,
    ) -> Union[Hunyuan3DFlowMatchSchedulerOutput, Tuple]:
        """Predict the sample from the previous timestep."""
        if isinstance(timestep, (int, torch.IntTensor, torch.LongTensor)):
            raise ValueError(
                "Passing integer indices as timesteps is not supported. "
                "Pass one of `scheduler.timesteps` as a timestep."
            )

        if self.step_index is None:
            self._init_step_index(timestep)

        # Upcast to avoid precision issues
        sample = sample.to(torch.float32)

        sigma = self.sigmas[self.step_index]
        sigma_next = self.sigmas[self.step_index + 1]

        prev_sample = sample + (sigma_next - sigma) * model_output
        prev_sample = prev_sample.to(model_output.dtype)

        self._step_index += 1

View on GitHub (pinned to 0132848349)

Solutions

1. Iterate directly over scheduler.timesteps and pass the raw float value
2. Convert with float(t) / t.float() if the value passed through numpy or casting
3. Never pass enumerate()'s index as the timestep

Example fix

# before
for i, t in enumerate(scheduler.timesteps):
    sample = scheduler.step(model_output, i, sample).prev_sample
# after
for t in scheduler.timesteps:
    sample = scheduler.step(model_output, t, sample).prev_sample

t = scheduler.timesteps[i]
assert not isinstance(t, (int, torch.IntTensor, torch.LongTensor))

def is_float_timestep(t) -> bool:
    return not isinstance(t, (int, torch.IntTensor, torch.LongTensor))

When it happens

Trigger: Looping `for i, t in enumerate(scheduler.timesteps)` and passing `t` that got cast to int, or passing the loop index i instead of t; passing timestep=int(t).

Common situations: Porting loop code from DDPM-style schedulers that accept ints; timesteps stored as numpy int64; tqdm progress loops using indices.

Related errors

• pair_postprocess must return a torch.Tensor
• Passing integer indices as timesteps is not supported.
• Passing integer indices (e.g. from `enumerate(timesteps)`) a
• Expected scheduler.sigmas to be a tensor for JoyEcho.
• Unsupported type {type(data)}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/806e30a70a945890. Report an issue: GitHub.