mudler/LocalAI · error

router/rerank: reranker is required (configure router.classi

Error message

router/rerank: reranker is required (configure router.classifier_model)

What it means

Construction-time panic in NewRerankClassifier: policies were supplied but the reranker backend handle is nil. The reranker (configured via router.classifier_model) is the cross-encoder the classifier delegates scoring to; without it there is no way to rank policies, so construction aborts.

Source

Thrown at core/services/routing/router/rerank.go:44

	// budget trims the query to the reranker model's context minus the
	// longest policy description (paired with the query per rerank call);
	// nil reranks Probe.Prompt as built by the caller.
	budget *lazyBudget
}

// defaultRerankActivationThreshold is the relevance floor a label
// must clear to be considered active. Reranker scores live in [0, 1]
// for cross-encoder / ColBERT heads; 0.5 picks "more positive than
// not on this label."
const defaultRerankActivationThreshold = 0.5

func NewRerankClassifier(policies []ScorePolicy, reranker backend.Reranker, cacheCap int, activationThreshold float64) *RerankClassifier {
	if len(policies) == 0 {
		panic("router/rerank: at least one policy is required")
	}
	if reranker == nil {
		panic("router/rerank: reranker is required (configure router.classifier_model)")
	}
	for _, p := range policies {
		if p.Label == "" {
			panic("router/rerank: policy has empty label")
		}
		if p.Description == "" {
			panic(fmt.Sprintf("router/rerank: policy %q has no description", p.Label))
		}
	}
	if activationThreshold <= 0 {
		activationThreshold = defaultRerankActivationThreshold
	}
	labels := make([]string, len(policies))
	docs := make([]string, len(policies))
	for i, p := range policies {
		labels[i] = p.Label
		docs[i] = p.Description
	}

View on GitHub (pinned to 44413a9d06)

Solutions

1. Set router.classifier_model in config to a model that provides rerank embeddings/scores (e.g. a cross-encoder)
2. Verify the named model exists and loaded successfully (check model list / logs at startup)
3. Ensure the referenced backend implements backend.Reranker (a rerank-capable model, not a plain LLM)
4. Fix construction order so the reranker handle is resolved before the router is assembled

Example fix

# before (yaml)
router:
  classifier:
    type: rerank
    policies: [...]
# after
router:
  classifier_model: rerank- ms-marco-minilm  # model id with rerank capability
  classifier:
    type: rerank
    policies: [...]

reranker, err := resolveModel(cfg.Router.ClassifierModel)
if err != nil || reranker == nil {
    return fmt.Errorf("router: classifier_model %q not loaded", cfg.Router.ClassifierModel)
}

func implementsReranker(m backend.Backend) bool { _, ok := m.(backend.Reranker); return ok }

When it happens

Trigger: Router config sets a rerank classifier with policies but router.classifier_model is unset/empty, or the named model failed to load so the loader passed nil; calling NewRerankClassifier(policies, nil, ...) directly.

Common situations: Enabling the classifier before pulling/installing the reranker model; classifier_model name typo so model resolution returns nil; the classifier model backend lacking rerank capability; ordering issue where the router is built before models are registered.

Related errors

• router/score: scorer is required (configure router.classifie
• router/rerank: at least one policy is required
• router/rerank: policy has empty label
• router/rerank: policy %q has no description
• router/score: at least one policy is required

AI-assisted analysis of mudler/LocalAI@44413a9d06 (2026-08-15). Data as JSON: /api/errors/e89823c25e299be3. Report an issue: GitHub.