mudler/LocalAI · error · ValueError

Unknown builtin reward function '{name}'. Available: {availa

Error message

Unknown builtin reward function '{name}'. Available: {available}

What it means

Reward specs with type 'builtin' must name a function in BUILTIN_REGISTRY (format_reward, reasoning_accuracy_reward, length_reward, etc.). Unknown names get a ValueError enumerating the available builtins, which acts as discoverable inline documentation of the registry.

Source

Thrown at backend/python/trl/reward_functions.py:222

    """
    if isinstance(specs_json, str):
        specs = json.loads(specs_json)
    else:
        specs = specs_json

    if not isinstance(specs, list):
        raise ValueError("reward_funcs must be a JSON array of reward function specs")

    reward_funcs = []
    for spec in specs:
        spec_type = spec.get("type", "builtin")
        name = spec.get("name", "")
        params = spec.get("params", {})

        if spec_type == "builtin":
            if name not in BUILTIN_REGISTRY:
                available = ", ".join(sorted(BUILTIN_REGISTRY.keys()))
                raise ValueError(
                    f"Unknown builtin reward function '{name}'. Available: {available}"
                )
            func = BUILTIN_REGISTRY[name]
            if params:
                # Convert string params to appropriate types
                typed_params = {}
                for k, v in params.items():
                    try:
                        typed_params[k] = int(v)
                    except (ValueError, TypeError):
                        try:
                            typed_params[k] = float(v)
                        except (ValueError, TypeError):
                            typed_params[k] = v
                func = functools.partial(func, **typed_params)
            reward_funcs.append(func)

        elif spec_type == "inline":

View on GitHub (pinned to 44413a9d06)

Solutions

1. Read the 'Available: ...' list in the error message and use one of those exact names.
2. Fix casing/underscores to match the registry (e.g. 'format_reward').
3. If you truly need custom logic, switch the spec to type 'inline' (requires the opt-in env var) instead of a nonexistent builtin.

Example fix

# before
{"type": "builtin", "name": "format-reward"}
# after
{"type": "builtin", "name": "format_reward"}

BUILTIN_NAMES = {"format_reward", "reasoning_accuracy_reward", "length_reward"}  # sync with registry

def builtin_known(name: str) -> bool:
    return name in BUILTIN_NAMES

try:
    build_reward_functions(specs)
except ValueError as e:
    if "Available:" in str(e):
        # parse the names from the message and suggest the closest
        raise ConfigError(str(e)) from e
    raise

When it happens

Trigger: spec name='accuracy' (not registered); typos like 'format-reward' or 'Format_Reward'; referencing a builtin added in a newer backend version.

Common situations: Guessing builtin names instead of checking the registry; version skew between backend and client docs; renaming drift after upgrading.

Related errors

• reward_funcs must be a JSON array of reward function specs
• Inline reward function '{name}' has no code
• Unknown reward function type '{spec_type}'. Use 'builtin' or
• GRPO requires at least one reward function. Specify reward_f
• Unsupported training method: {training_method}. Supported: s

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/9d9ae0d02434fd68. Report an issue: GitHub.