sgl-project/sglang · critical · ValueError

{selection_error}{component_suffix}

Error message

{selection_error}{component_suffix}

What it means

Raised by get_attn_backend in the multimodal runtime's attention backend selector when a candidate backend was tried and its own selection logic raised an error (e.g. the backend's constructor/requirements check failed). The original exception is chained via `from selection_error`, so the cause message is embedded at the front of the string, followed by a component suffix naming the attention component.

Source

Thrown at python/sglang/multimodal_gen/runtime/layers/attention/selector.py:291

            continue

        attention_backend_cls = candidate_cls
        if candidate_index > 0:
            fallback_reason = allowed_fallback_reason
        break

    if attention_backend_cls is None:
        component_name = get_component_attn_backend_name()
        component_suffix = (
            f" for component '{component_name}'" if component_name is not None else ""
        )
        if unsupported_requirements:
            raise ValueError(
                f"Attention backend '{unsupported_backend_name}' does not implement "
                f"{', '.join(unsupported_requirements)}{component_suffix}"
            )
        if selection_error is not None:
            raise ValueError(
                f"{selection_error}{component_suffix}"
            ) from selection_error
        raise ValueError(
            f"No compatible attention backend is available{component_suffix}"
        )

    backend_name = attention_backend_cls.get_enum().name.lower()
    reason = fallback_reason
    if reason is None and backend_name == constraint_backend:
        reason = "component constraint"
    if not _record_component_attn_backend(backend_name, reason):
        reason_suffix = f" ({reason})" if reason else ""
        logger.info_once(f"Using {backend_name} attention backend{reason_suffix}")
    return attention_backend_cls


@cache
def _cached_get_attn_backend(

View on GitHub (pinned to 0132848349)

Solutions

1. Read the chained cause (`__cause__`) — the real failure is the selection_error text, fix that (install backend deps, adjust head_size/dtype/config)
2. Verify the attention backend name in the model/server config is valid for your platform
3. If the underlying error is a missing optional dependency, install it (e.g. flashinfer) and retry

Example fix

# before: server fails with wrapped selection error
# after: inspect cause and fix root issue
try:
    get_attn_backend(...)
except ValueError as e:
    logger.error("root cause: %s", e.__cause__)
    raise

from sglang.multimodal_gen.runtime.layers.attention.selector import get_attn_backend
try:
    get_attn_backend(name, head_size, dtype, supported_backends)
except ValueError:
    ...

try:\n    backend = get_attn_backend(...)\nexcept ValueError as e:\n    root = e.__cause__ or e\n    logger.error("backend selection failed: %s", root)\n    raise

When it happens

Trigger: Calling get_attn_backend (directly or via layer __init__ / prepare_attention_backend_override) where every candidate backend's internal selection raises; the selection_error string is prepended to the message.

Common situations: A model config requests an attention backend whose __init__ fails on the current platform (e.g. flashinfer missing, unsupported dtype/head size), and the selector falls through to re-raising the underlying error wrapped in ValueError.

Related errors

• No compatible attention backend is available{component_suffi
• For multimodal input processing do not set `enable_tokenizer
• Sparse Video Gen 2 attention does not support causal attenti
• Attention backend '{selected_backend}' is not supported by t
• f"Unsupported patch_size type: {type(patch_size)}"

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/3c7c21e2ae9f8df2. Report an issue: GitHub.