mudler/LocalAI · error · ValueError

Unknown reward function type '{spec_type}'. Use 'builtin' or

Error message

Unknown reward function type '{spec_type}'. Use 'builtin' or 'inline'

What it means

Each reward spec's 'type' field must be exactly 'builtin' or 'inline'; anything else ('function', 'custom', 'python', or a missing type that is not the defaulted 'builtin'... missing type defaults to builtin, so this fires only on explicitly wrong values) reaches the else branch and raises with the accepted values.

Source

Thrown at backend/python/trl/reward_functions.py:256

            reward_funcs.append(func)

        elif spec_type == "inline":
            if not _inline_rewards_allowed():
                raise ValueError(
                    f"Inline reward function '{name}' rejected: inline reward code "
                    f"executes arbitrary Python and is disabled by default. Set "
                    f"{ALLOW_INLINE_ENV}=true on the backend to enable it (only on a "
                    f"trusted, access-controlled instance), or use a builtin reward "
                    f"function instead."
                )
            code = spec.get("code", "")
            if not code.strip():
                raise ValueError(f"Inline reward function '{name}' has no code")
            func = compile_inline_reward(name, code)
            reward_funcs.append(func)

        else:
            raise ValueError(f"Unknown reward function type '{spec_type}'. Use 'builtin' or 'inline'")

    return reward_funcs

View on GitHub (pinned to 44413a9d06)

Solutions

1. Set type to 'builtin' for registry functions or 'inline' for code (with the env opt-in).
2. Omit 'type' entirely when you mean builtin — it defaults to that.
3. Upgrade the backend if you expected a newer spec kind.

Example fix

# before
{"type": "custom", "name": "format_reward"}
# after
{"type": "builtin", "name": "format_reward"}

VALID_TYPES = {"builtin", "inline"}

def spec_type_valid(spec: dict) -> bool:
    return spec.get("type", "builtin") in VALID_TYPES

def is_valid_reward_spec(spec) -> bool:
    return (
        isinstance(spec, dict)
        and spec.get("type", "builtin") in {"builtin", "inline"}
        and isinstance(spec.get("name", ""), str)
    )

try:
    build_reward_functions(specs)
except ValueError as e:
    if "Unknown reward function type" in str(e):
        for s in specs:
            s.setdefault("type", "builtin")
        retry()
    raise

When it happens

Trigger: spec = {'type':'custom','name':...}; 'type':'BuiltIn' with wrong case; future spec kinds not understood by this backend version.

Common situations: Clients inventing new spec types; schema drift after an API update; copy-paste from docs describing a newer version.

Related errors

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

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/9c23b43d8688775b. Report an issue: GitHub.