sgl-project/sglang · error · ValueError

task {task!r} does not belong to partition {partition!r}

Error message

task {task!r} does not belong to partition {partition!r}

What it means

Each declared task must belong to the partition this metadata describes: partition_for_task(task) must equal the metadata's partition. Mixing tasks from another weight partition into this metadata block raises this error at load time.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/release_metadata.py:107

            raise ValueError(
                "model_index.json._minimax_h3.sigma_shift_scales requires numeric "
                "video and audio values"
            ) from exc
        metadata = cls(
            schema_version=1,
            partition=partition,
            tasks=tasks,
            task_aliases=dict(aliases),
            video_sigma_shift=video_sigma,
            audio_sigma_shift=audio_sigma,
        )
        for task in metadata.tasks:
            if canonical_minimax_h3_task(task) != task:
                raise ValueError(
                    f"tasks must contain canonical task names, got {task!r}"
                )
            if partition_for_task(task) != partition:
                raise ValueError(
                    f"task {task!r} does not belong to partition {partition!r}"
                )
        for alias, target in metadata.task_aliases.items():
            if target not in metadata.tasks:
                raise ValueError(
                    f"task alias {alias!r} targets undeclared task {target!r}"
                )
            if canonical_minimax_h3_task(alias) != target:
                raise ValueError(
                    f"unsupported task alias mapping {alias!r} -> {target!r}"
                )
        return metadata

    @property
    def sigma_shift_scales(self) -> dict[str, float]:
        return {"video": self.video_sigma_shift, "audio": self.audio_sigma_shift}

    def canonical_task(self, task: str) -> str:

View on GitHub (pinned to 0132848349)

Solutions

1. Check partition_for_task for each listed task and remove tasks that resolve to a different partition
2. Confirm the partition field of this metadata block matches the checkpoint actually being loaded
3. Split the tasks into the correct per-partition model_index.json files

Example fix

// before
{"partition": "video", "tasks": ["text_to_video", "text_to_audio"]}
// after
{"partition": "video", "tasks": ["text_to_video"]}

assert all(partition_for_task(t) == partition for t in tasks), "task belongs to another partition"

When it happens

Trigger: A model_index.json for partition "a" that lists a task whose partition_for_task resolves to "b", e.g. an audio task inside a video-partition metadata block.

Common situations: Merging metadata blocks when combining checkpoints, or copying the tasks list from a different partition's index during model packaging.

Understand the failure class

Background: Schema validation failed / invalid input schema: payload rejected because its shape doesn't match the expected schema — this error's family across 28 libraries.

Related errors

• task {task!r} resolves outside partition {self.partition!r}
• model_index.json._minimax_h3.partition must be one of fl2va,
• task {task!r} is not served by MiniMax H3 partition {self.pa
• {path} must be a non-empty list
• {path} must contain non-empty strings

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ff5d061dc322af99. Report an issue: GitHub.