BerriAI/litellm · error · ValueError

quality_tier={prefs.quality_tier} is not supported; valid ti

Error message

quality_tier={prefs.quality_tier} is not supported; valid tiers are {valid}

What it means

Config validation in the adaptive-router bandit priors: the requested quality_tier in AdaptiveRouterPreferences is not one of the valid tier values (listed in the message), so no base prior weight exists for it.

Source

Thrown at litellm/router_strategy/adaptive_router/bandit.py:55

        total: Final = self.alpha + self.beta
        return self.alpha / total if total > 0 else 0.5

    @property
    def total_samples(self) -> int:
        return max(0, int(self.alpha + self.beta - COLD_START_MASS))


def initial_cell(prefs: AdaptiveRouterPreferences, request_type: RequestType) -> BanditCell:
    """
    Cold-start prior for a (model, request_type) cell.

    mean = base_tier_weight[tier] + (STRENGTH_BONUS if request_type in strengths else 0)
    capped at 0.95 to avoid an over-confident prior.
    Total mass = COLD_START_MASS so that ~10 real observations can move it noticeably.
    """
    if prefs.quality_tier not in BASE_TIER_WEIGHT:
        valid: Final = sorted(BASE_TIER_WEIGHT)
        raise ValueError(f"quality_tier={prefs.quality_tier} is not supported; valid tiers are {valid}")
    base: Final = BASE_TIER_WEIGHT[prefs.quality_tier]
    bonus: Final = STRENGTH_BONUS if request_type in prefs.strengths else 0.0
    mean: Final = min(0.95, base + bonus)
    alpha: Final = mean * COLD_START_MASS
    beta: Final = (1.0 - mean) * COLD_START_MASS
    return BanditCell(alpha=alpha, beta=beta)


def apply_delta(cell: BanditCell, delta_alpha: float, delta_beta: float) -> BanditCell:
    """
    Apply a learning update to a cell, enforcing the sample cap.

    SAMPLE_CAP is a HARD cap on (alpha + beta). When the cap would be exceeded,
    we drop the update. (D5: hard cap, no rescaling — keep v0 simple.)
    """
    new_alpha: Final = cell.alpha + delta_alpha
    new_beta: Final = cell.beta + delta_beta
    if new_alpha + new_beta > SAMPLE_CAP:

View on GitHub (pinned to 77b7c6c40c)

Solutions

1. Use one of the listed valid quality tiers in routing preferences.

When it happens

Trigger: Thrown at litellm/router_strategy/adaptive_router/bandit.py:55 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of BerriAI/litellm@77b7c6c40c (2026-08-18). Data as JSON: /api/errors/8522695bc8b08f3a. Report an issue: GitHub.