sgl-project/sglang · error · ValueError

Ambiguous tensor payload for multi-module update. Provide a

Error message

Ambiguous tensor payload for multi-module update. Provide a dict mapping module_name -> module payload, requested modules: {module_names}.

What it means

Raised by WeightsUpdater._resolve_module_payloads when a non-dict payload is given but the update targets more than one module. With multiple modules the updater cannot tell which payload belongs to which module, so it demands a dict keyed by module name.

Source

Thrown at python/sglang/multimodal_gen/runtime/post_training/weights_updater.py:759

    def _resolve_module_payloads(
        self,
        named_tensors: Any,
        modules_to_update: list[tuple[str, torch.nn.Module]],
    ) -> dict[str, Any]:
        module_names = [name for name, _ in modules_to_update]
        if isinstance(named_tensors, dict):
            missing = [name for name in module_names if name not in named_tensors]
            if missing:
                raise ValueError(
                    f"Missing tensor payload for module(s): {missing}. "
                    f"Provided modules: {list(named_tensors.keys())}"
                )
            return {name: named_tensors[name] for name in module_names}

        if len(module_names) == 1:
            return {module_names[0]: named_tensors}

        raise ValueError(
            "Ambiguous tensor payload for multi-module update. "
            "Provide a dict mapping module_name -> module payload, "
            f"requested modules: {module_names}."
        )

    def _materialize_weights_iter(self, module_payload: Any, load_format: str | None):
        if load_format == "flattened_bucket":
            if not isinstance(module_payload, dict):
                raise ValueError(
                    "flattened_bucket payload must be a dict with "
                    "'flattened_tensor' and 'metadata'."
                )
            flattened_tensor = module_payload.get("flattened_tensor")
            metadata = module_payload.get("metadata")
            if flattened_tensor is None or metadata is None:
                raise ValueError(
                    "flattened_bucket payload missing 'flattened_tensor' or 'metadata'."
                )

View on GitHub (pinned to 0132848349)

Solutions

1. Pass a dict {module_name: payload} covering every requested module
2. Or set target_modules to exactly one module so the single payload can be assigned to it

Example fix

// before
updater.update_weights_from_tensor(named_tensors=flat_payload, target_modules=["a","b"])
// after
updater.update_weights_from_tensor(named_tensors={"a": payload_a, "b": payload_b})

if len(target_modules) > 1 and not isinstance(named_tensors, dict):
    named_tensors = {target_modules[0]: named_tensors}  # only valid if you truly want one module
# otherwise build a per-module dict

def is_multi_module_payload(named_tensors: Any, n: int) -> bool:
    return n == 1 or isinstance(named_tensors, dict)

When it happens

Trigger: Calling update_weights_from_tensor with named_tensors being a single payload object (list/tuple/flattened bucket) while target_modules (or the default full module set) contains 2+ modules. A single payload is only accepted when exactly one module is being updated.

Common situations: Code originally written for a single-module pipeline reused on a multi-module pipeline; defaulting target_modules=None (all modules) but passing one flat payload.

Related errors

• Missing tensor payload for module(s): {missing}. Provided mo
• Module(s) requested for update not found in pipeline: {unkno
• flattened_bucket payload must be a dict with 'flattened_tens
• flattened_bucket payload missing 'flattened_tensor' or 'meta
• Unsupported module payload type for load_format={load_format

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/4d26c37845a4e6d8. Report an issue: GitHub.