sgl-project/sglang · error · ValueError

No processor registered for architecture

Error message

No processor registered for architecture: {hf_config.architectures}.
Registered architectures: {[model_cls.__name__ for model_cls in PROCESSOR_MAPPING.keys()]}

What it means

get_mm_processor looks up the model architecture in PROCESSOR_MAPPING; an architecture with no registered multimodal processor means SGLang has no MM preprocessing pipeline for that model, so serving it (especially with MM inputs or the rust path) fails.

Solutions

Upgrade SGLang to a version supporting the architecture
Check PROCESSOR_MAPPING keys and register a processor for your architecture
Fix the model's config.json architectures to match a supported canonical name
Use a supported checkpoint of the same model family

Defensive patterns

Strategy: fallback

Validate before calling

from sglang.srt.managers.multimodal_processor import PROCESSOR_MAPPING
archs = model_config.hf_config.architectures
assert any(a in PROCESSOR_MAPPING for a in archs), f'no MM processor for {archs}'

Try / catch

try:
    get_mm_processor(...)
except ValueError as e:
    if 'No processor registered' in str(e):
        raise SystemExit('upgrade sglang or use a supported checkpoint')
    raise

Prevention

Pin/upgrade SGLang to a release covering your VLM
Verify architectures in config.json before deploying

When it happens

Trigger: Loading a new/unsupported VLM whose hf_config.architectures entry is absent from PROCESSOR_MAPPING; custom model classes with renamed architectures.

Common situations: Newly released VLM on an older SGLang version; community checkpoint with non-standard architecture name; typo in config.json architectures.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/14920e3c393f2788. Report an issue: GitHub.

Appendix: source

Thrown at python/sglang/srt/managers/multimodal_processor.py:84

            TransformersAutoMultimodalProcessor,
        )

        return TransformersAutoMultimodalProcessor

    return None


def get_mm_processor(
    hf_config,
    server_args: ServerArgs,
    processor,
    transport_mode,
    model_config=None,
    **kwargs,
) -> BaseMultimodalProcessor:
    processor_cls = get_mm_processor_cls(hf_config, server_args, model_config)
    if processor_cls is None:
        raise ValueError(
            f"No processor registered for architecture: {hf_config.architectures}.\n"
            f"Registered architectures: {[model_cls.__name__ for model_cls in PROCESSOR_MAPPING.keys()]}"
        )
    return processor_cls(hf_config, server_args, processor, transport_mode, **kwargs)

View on GitHub (pinned to 0132848349)