sgl-project/sglang · error · ValueError

Timesteps must be provided

Error message

Timesteps must be provided

What it means

The Hunyuan3D denoising loop stage expects timesteps to have been computed by an earlier scheduler-setup stage and carried on the batch object. If batch.timesteps is None the loop cannot run because the diffusion schedule is undefined. This indicates a broken or skipped pipeline stage ordering rather than a user-facing parameter mistake.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/hunyuan3d/shape.py:324

        """Prepare Hunyuan3D-specific variables for the base denoising loop."""
        assert self.transformer is not None
        pipeline = self.pipeline() if self.pipeline else None
        scheduler = batch.scheduler
        assert scheduler is not None
        cache_dit_num_inference_steps = batch.extra.get(
            "cache_dit_num_inference_steps", batch.num_inference_steps
        )
        freshly_loaded = load_transformer_if_needed(self, server_args)
        if freshly_loaded:
            self._maybe_enable_cache_dit(cache_dit_num_inference_steps, batch)
            self._maybe_torch_compile(self.transformer)
            register_loaded_transformer(self, server_args, pipeline)
        else:
            self._maybe_enable_cache_dit(cache_dit_num_inference_steps, batch)

        timesteps = batch.timesteps
        if timesteps is None:
            raise ValueError("Timesteps must be provided")

        latents = batch.latents
        if latents is None:
            raise ValueError("Latents must be provided")

        cond = batch.prompt_embeds[0] if batch.prompt_embeds else None
        if cond is None:
            raise ValueError("Conditioning (prompt_embeds) must be provided")

        if batch.raw_latent_shape is None:
            batch.raw_latent_shape = latents.shape

        guidance = batch.extra.get("shape_guidance")
        num_inference_steps = batch.num_inference_steps
        num_warmup_steps = len(timesteps) - num_inference_steps * scheduler.order

        extra_step_kwargs = self.prepare_extra_func_kwargs(
            scheduler.step,

View on GitHub (pinned to 0132848349)

Solutions

1. Ensure the pipeline includes and executes the scheduler/timestep-setup stage before the Hunyuan3D denoising stage
2. Inspect pipeline composition/logs to confirm the upstream stage ran and attached timesteps to the batch
3. When constructing batches manually (tests), populate batch.timesteps from the configured scheduler

Example fix

# before
batch = Req(latents=z, prompt_embeds=[c])  # no timesteps
stage.forward(batch, server_args)

# after
batch.timesteps = scheduler.set_timesteps(num_inference_steps).timesteps
stage.forward(batch, server_args)

if batch.timesteps is None:
    batch.timesteps = scheduler.set_timesteps(batch.num_inference_steps).timesteps

def has_timesteps(batch) -> bool:
    return getattr(batch, "timesteps", None) is not None

When it happens

Trigger: Running the shape denoising stage without its preceding scheduler/timestep-computation stage, or when that upstream stage failed silently / was disabled in the pipeline composition. Also when constructing the batch manually for tests and forgetting timesteps.

Common situations: Custom or reordered pipelines that omit the scheduler stage; warmup or test harnesses building Req objects by hand; a refactor that renamed the field carrying timesteps; upstream stage crashed before populating the batch.

Related errors

• Latents must be provided
• Conditioning (prompt_embeds) must be provided
• SANA-WM denoising requires prepared timesteps.
• {response.error}
• action policy returned no output

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/5676c0e0ebfa32ca. Report an issue: GitHub.