sgl-project/sglang · error · ValueError

Could not parse attention backend config: {config_str}

Error message

Could not parse attention backend config: {config_str}

What it means

The attention-backend config string (e.g. --attention-backend flash:kv_cache_dtype=fp8,...) is parsed key-by-key into bool/int/float/str. If any part of parsing or value coercion throws, the generic 'Could not parse attention backend config' error wraps it, discarding the underlying cause.

Source

Thrown at python/sglang/multimodal_gen/runtime/server_args/server_args.py:1724

        # 3. treat as k=v pairs (simple implementation). e.g., "sparsity=0.5,enable_x=true"
        try:
            config = {}
            pairs = config_str.split(",")
            for pair in pairs:
                k, v = pair.split("=", 1)
                k = k.strip()
                v = v.strip()
                if v.lower() == "true":
                    v = True
                elif v.lower() == "false":
                    v = False
                elif v.replace(".", "", 1).isdigit():
                    v = float(v) if "." in v else int(v)
                config[k] = v
            return config
        except Exception:
            raise ValueError(f"Could not parse attention backend config: {config_str}")

    def __post_init__(self):
        # configure logger before use
        configure_logger(server_args=self)

        component_paths: dict[str, str] = {}
        component_weights_paths = dict(self.component_weights_paths)
        for component, path in self.component_paths.items():
            supports_weight_file_override = (
                is_dit_component_name(component)
                or is_text_encoder_component_name(component)
                or is_image_encoder_component_name(component)
                or is_vae_component_name(component)
            )
            if (
                not supports_weight_file_override
                or not is_explicit_weight_file_reference(path)
            ):

View on GitHub (pinned to 0132848349)

Solutions

1. Simplify the config string to confirm the backend name alone parses, then add one key=value at a time to find the bad token
2. Match the documented 'name:key=value,key=value' grammar exactly, quoting the whole flag in shell
3. Check for stray characters from shell expansion (unquoted colons/commas)

Example fix

# before
--attention-backend 'flash:kv_cache_dtype==fp8'
# after
--attention-backend 'flash:kv_cache_dtype=fp8'

def validate_backend_config(s: str) -> bool:
    try:
        name, _, rest = s.partition(':')
        return bool(name) and all('=' in kv for kv in rest.split(',') if kv)
    except Exception:
        return False

try:
    parsed = ServerArgs._parse_attention_backend_config(cfg)
except ValueError:
    logger.warning('bad attention backend config %r; using default', cfg)
    parsed = {}

When it happens

Trigger: Passing malformed syntax like 'flash:kv_cache_dtype=' or 'flash:flag==true', or an unterminated JSON-ish fragment; any non-numeric, non-bool, non-str token that breaks the coercion ladder.

Common situations: Hand-writing backend override strings; shell quoting mangling colons/commas; version changes in the accepted config grammar.

Related errors

• This browser cannot encode H.264 MP4
• H.264 encoder did not return MP4 decoder config
• delta payload size mismatch: expected ${expectedSize}, got $
• Sparse Video Gen 2 attention does not support causal attenti
• {selection_error}{component_suffix}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/417647aaf91035c0. Report an issue: GitHub.