sgl-project/sglang · error · AttributeError

Subclasses of BaseDiT must define '{attr}' class variable

Error message

Subclasses of BaseDiT must define '{attr}' class variable

What it means

BaseDiT uses __init_subclass__ to enforce that every DiT model implementation declares a set of required class-level attributes (e.g. param_names_mapping, _compile_conditions). Defining a subclass without these attributes raises AttributeError at class-definition time, i.e. at import of the module containing the subclass, not at instantiation.

Source

Thrown at python/sglang/multimodal_gen/runtime/models/dits/base.py:65

        AttentionBackendEnum.VIDEO_SPARSE_ATTN,
        AttentionBackendEnum.SPARSE_VIDEO_GEN_2_ATTN,
        AttentionBackendEnum.VMOBA_ATTN,
        AttentionBackendEnum.SAGE_ATTN_3,
        AttentionBackendEnum.LASER_ATTN,
        AttentionBackendEnum.BLOCK_SPARSE_ATTN,
        AttentionBackendEnum.RAIN_FUSION_ATTN,
    }

    def __init_subclass__(cls) -> None:
        required_class_attrs = [
            "_fsdp_shard_conditions",
            "param_names_mapping",
            "_compile_conditions",
        ]
        super().__init_subclass__()
        for attr in required_class_attrs:
            if not hasattr(cls, attr):
                raise AttributeError(
                    f"Subclasses of BaseDiT must define '{attr}' class variable"
                )

    def __init__(self, config: DiTConfig, hf_config: dict[str, Any], **kwargs) -> None:
        super().__init__()
        # `config.arch_config` contains static model metadata. Runtime
        # capabilities remain class attributes on the model implementation.
        self.config: DiTArchConfig = config.arch_config
        self.prefix = config.prefix
        self.hf_config = hf_config
        if not self.supported_attention_backends:
            raise ValueError(
                f"Subclass {self.__class__.__name__} must define _supported_attention_backends"
            )

    @abstractmethod
    def forward(
        self,

View on GitHub (pinned to 0132848349)

Solutions

1. Add the missing attribute(s) named in the message as class variables on your BaseDiT subclass, e.g. param_names_mapping = (...) and _compile_conditions = ...
2. Copy the full attribute set from an existing model implementation (e.g. cosmos3video.py or another dits/ model) and adapt values
3. Check base.py's required_class_attrs list to see the complete set of required names

Example fix

// before
class MyDiT(BaseDiT):
    ...
// after
class MyDiT(BaseDiT):
    param_names_mapping = {...}
    _compile_conditions = ...
    ...

from sglang.multimodal_gen.runtime.models.dits.base import BaseDiT
REQUIRED = ["param_names_mapping", "_compile_conditions"]
def check_dit_subclass(cls):
    missing = [a for a in REQUIRED if not hasattr(cls, a)]
    assert not missing, f"missing class attrs: {missing}"
    return True

def is_valid_dit_subclass(cls) -> bool:
    return isinstance(cls, type) and issubclass(cls, BaseDiT) and all(
        hasattr(cls, a) for a in ("param_names_mapping", "_compile_conditions")
    )

When it happens

Trigger: Creating any class that inherits from BaseDiT without defining all names in required_class_attrs (e.g. missing param_names_mapping or _compile_conditions). The error fires the moment the subclass statement is executed (import time).

Common situations: Adding a new DiT model to sglang's multimodal_gen runtime and forgetting boilerplate class attributes; copying an older model implementation that predates a newly-added required attribute after a version upgrade.

Related errors

• Subclass {self.__class__.__name__} must define _supported_at
• Subclasses of BaseDiT must define '{attr}' instance variable
• Unknown serve backend {name!r}. Available values: {available
• Multiple distributions register serve backend {name!r}: {pro
• Failed to load serve backend {name!r} from {self._entry_poin

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/0e8e36cd6ba18bde. Report an issue: GitHub.