sgl-project/sglang · error · ValueError

The quantization config must be a subclass of `QuantizationC

Error message

The quantization config must be a subclass of `QuantizationConfig`.

What it means

The register_quantization decorator requires the decorated class to be a subclass of QuantizationConfig. Passing an arbitrary class fails this isinstance check because the framework needs the QuantizationConfig interface (get_config_filenames, from_config, etc.) to serve the method.

Source

Thrown at python/sglang/multimodal_gen/runtime/layers/quantization/__init__.py:82

def register_quantization_config(quantization: str):
    """Register a customized vllm quantization config.

    When a quantization method is not supported by vllm, you can register a customized
    quantization config to support it.

    Args:
        quantization (str): The quantization method name.


    """  # noqa: E501

    def _wrapper(quant_config_cls):
        if quantization in QUANTIZATION_METHODS:
            raise ValueError(
                f"The quantization method `{quantization}` is already exists."
            )
        if not issubclass(quant_config_cls, QuantizationConfig):
            raise ValueError(
                "The quantization config must be a subclass of " "`QuantizationConfig`."
            )
        _CUSTOMIZED_METHOD_TO_QUANT_CONFIG[quantization] = quant_config_cls
        QUANTIZATION_METHODS.append(quantization)
        return quant_config_cls

    return _wrapper


def get_quantization_config(quantization: str) -> type[QuantizationConfig]:
    if quantization not in QUANTIZATION_METHODS:
        raise ValueError(f"Invalid quantization method: {quantization}")

    method_to_config: dict[str, type[QuantizationConfig]] = {}
    # Update the `method_to_config` with customized quantization methods.
    method_to_config.update(_CUSTOMIZED_METHOD_TO_QUANT_CONFIG)

    return method_to_config[quantization]

View on GitHub (pinned to 0132848349)

Solutions

1. Make your class inherit from QuantizationConfig (directly or via an existing subclass)
2. Implement the required abstract methods such as from_config and get_config_filenames
3. If you meant to register an existing config type, decorate the actual QuantizationConfig subclass

Example fix

// before
@register_quantization("my_quant")
class MyConfig:  # not a QuantizationConfig
    ...

// after
from sglang...quantization import QuantizationConfig

@register_quantization("my_quant")
class MyConfig(QuantizationConfig):
    ...

def is_quant_config(cls: type) -> bool:
    import inspect
    from ...quantization import QuantizationConfig
    return inspect.isclass(cls) and issubclass(cls, QuantizationConfig)

When it happens

Trigger: Decorating a class that does not inherit from QuantizationConfig with @register_quantization(...), e.g. a plain dataclass or a subclass of a different config base.

Common situations: Writing a custom quant backend without reading the base-class contract; refactoring a config class so it accidentally loses its QuantizationConfig parent; multiple inheritance ordering issues.

Related errors

• rollout_noise_level must be a number, got {noise!r}
• The quantization method `{quantization}` is already exists.
• Invalid quantization method: {quantization}
• kitchen_w4a4 is inferred from per-layer checkpoint metadata;
• GGUFConfig must be constructed from a GGUF checkpoint

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/4ce45188a4740c3c. Report an issue: GitHub.