sgl-project/sglang · error · ValueError

Invalid {field_name}={value!r}.

Error message

Invalid {field_name}={value!r}.

What it means

Raised by _parse_optional_int while parsing the DFLASH draft config: the field's value is not None and cannot be converted with int(). This is a config deserialization error — a required-numeric field was given a non-numeric value such as a bad string or a nested object.

Source

Thrown at python/sglang/srt/speculative/dflash_utils.py:519

    sample_from_anchor = dflash_config.get(
        "sample_from_anchor", _cfg_get(config, "sample_from_anchor", True)
    )
    return sample_from_anchor is False


def _parse_optional_int(
    value: Any,
    *,
    field_name: str,
    min_value: Optional[int] = None,
) -> Optional[int]:
    if value is None:
        return None
    try:
        parsed = int(value)
    except Exception as e:
        raise ValueError(f"Invalid {field_name}={value!r}.") from e
    if min_value is not None and parsed < int(min_value):
        comparator = "positive" if int(min_value) == 1 else f">= {int(min_value)}"
        raise ValueError(f"{field_name} must be {comparator}, got {parsed}.")
    return parsed


@dataclass(frozen=True)
class DFlashDraftConfig:
    num_hidden_layers: Optional[int]
    num_target_layers: Optional[int]
    block_size: Optional[int]
    conv_kernel_size: int
    conv_group_size: int
    selector_rank: int
    selector_top_k: int
    output_multiplier: float
    final_logit_softcapping: Optional[float]
    target_layer_ids: Optional[List[int]]

View on GitHub (pinned to 0132848349)

Solutions

1. Fix the offending field in the draft config to be an integer or null.
2. Use the error's chained cause (raise ... from e) and the field_name in the message to identify which field failed, then correct just that entry.
3. Validate the draft config JSON with a schema/lint before deploying.

Example fix

# before
{"speculative_num_draft_tokens": "eight"}
# after
{"speculative_num_draft_tokens": 8}

for k, v in draft_cfg.items():
    if v is not None and not isinstance(v, (int, float)):
        try:
            int(v)
        except (TypeError, ValueError):
            raise ValueError(f'{k} must be int-convertible, got {v!r}')

def is_int_like(v) -> bool:
    if v is None or isinstance(v, bool):
        return v is None
    try:
        int(v)
        return True
    except (TypeError, ValueError):
        return False

try:
    cfg = parse_dflash_draft_config(raw)
except ValueError as e:
    raise ConfigError(f'draft config invalid: {e}') from e

When it happens

Trigger: parse_dflash_draft_config on a config where a field like speculative_num_draft_tokens is "eight", "", [8], or a dict; any int(field) call raising ValueError/TypeError.

Common situations: Hand-edited draft config JSON with quoted or malformed numbers; YAML/JSON passing a list where a scalar int is expected; env-var overrides injected as strings like "null" instead of null.

Related errors

• DFLASH config.layer_types must be a sequence of attention ty
• {option} must be a dict or a comma-separated component=value
• DFLASH speculative decoding only supports CUDA and NPU devic
• Currently DFLASH speculative decoding does not support dp at
• Currently DFLASH speculative decoding only supports pp_size

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/26d0ddb5a594bcf4. Report an issue: GitHub.