CherryHQ/cherry-studio · error · Error

Rerank response results must reference a valid document inde

Error message

Rerank response results must reference a valid document index

What it means

Re-thrown during Vertex AI model listing when options.throwOnError is set and at least one publisher group failed (returned null). `firstPublisherError` is captured as `unknown` from the per-publisher catch; if it is not an Error instance but is defined (e.g. a thrown string, a fetch rejection wrapping a non-Error), it is stringified and re-wrapped so the original value reaches the caller as a message. This branch only runs when the underlying thrown value was a non-Error.

Source

Thrown at packages/ai-sdk-provider/src/openai-compatible-reranking-model.ts:101

}

function parseRerankResponse(body: unknown, documentCount: number): RerankRanking {
  const results = (body as OpenAICompatibleRerankResponse).results
  if (!Array.isArray(results)) {
    throw new Error('Rerank response must contain a results array')
  }

  return results.map((result) => {
    if (typeof result !== 'object' || result === null) {
      throw new Error('Rerank response results must be objects')
    }

    if (typeof result.index !== 'number' || typeof result.relevance_score !== 'number') {
      throw new Error('Rerank response results must contain numeric index and relevance_score')
    }

    if (!Number.isInteger(result.index) || result.index < 0 || result.index >= documentCount) {
      throw new Error('Rerank response results must reference a valid document index')
    }

    return { index: result.index, relevanceScore: result.relevance_score }
  })
}

export function createOpenAICompatibleRerankingModel(
  modelId: string,
  settings: OpenAICompatibleRerankingModelSettings
): RerankingModelV3 {
  const baseURL = withoutTrailingSlash(settings.baseURL)
  if (!baseURL) {
    throw new Error('OpenAI-compatible reranking model requires baseURL')
  }

  return new OpenAICompatibleRerankingModel(modelId, {
    provider: `${settings.name}.rerank`,
    url: ({ path }) => {

View on GitHub (pinned to 726446b54c)

Solutions

1. The stringified message is the underlying failure — read it: it usually names the auth or HTTP cause.
2. Verify the Vertex provider is configured with valid iam-gcp credentials (service account JSON + project + region).
3. Confirm the service account has the Vertex AI User / Model Garden access IAM role.
4. Retry once for transient 5xx; if one publisher consistently fails, note that non-throwOnError mode would skip it and list the rest.
5. Run the listing without throwOnError to see which publishers succeed and which fail individually in the warn logs.

Example fix

// before: caller passes throwOnError and gets an opaque String(...) message
await listModels(provider, signal, { throwOnError: true })
// after: list permissively, then inspect warn logs to find the failing publisher, fix its auth/region, then re-enable strict mode
const models = await listModels(provider, signal)  // throwOnError defaults off; failures degrade to 'no models from this publisher'
// ... fix the failing publisher's config ...
await listModels(provider, signal, { throwOnError: true })

// Validate Vertex auth before requesting a strict (throwOnError) listing.
const auth = providerService.getAuthConfig(provider.id)
if (auth?.type !== 'iam-gcp' || !auth.serviceAccountJson) {
  throw new Error('Vertex provider must use iam-gcp auth with a service account JSON')
}
await listModels(provider, signal, { throwOnError: true })

try {
  await listModels(provider, signal, { throwOnError: true })
} catch (e) {
  // the message stringifies the original (non-Error) cause; surface it, do not blanket-retry
  logger.error('Vertex listing failed', { providerId: provider.id, cause: e instanceof Error ? e.message : String(e) })
  // fall back to non-strict listing to still surface working publishers
  await listModels(provider, signal)
}

When it happens

Trigger: A per-publisher Vertex request (`publishers/{publisher}/models`) threw something that is not an Error instance — e.g. a raw string, a Response object, or a library that rejects with a plain object — and throwOnError propagated it. The real failure is upstream (auth, network, 403); this error is just the re-wrapping.

Common situations: Vertex service account JSON invalid or expired (GCE/gcloud auth); the configured region does not serve a publisher (e.g. Model Garden model unavailable in region); IAM permissions lacking `aiplatform.models.list`; a dependency rejecting with a non-Error value; transient 5xx from the Vertex endpoint.

Related errors

• Provider extension "${id}" not found. Did you forget to regi
• VertexAI requires iam-gcp auth configuration.
• OpenAI-compatible reranking model requires baseURL
• Private key must be a non-empty string
• Invalid PEM format: missing BEGIN/END markers or key content

AI-assisted analysis of CherryHQ/cherry-studio@726446b54c (2026-08-12). Data as JSON: /api/errors/298a8e30631e26c7. Report an issue: GitHub.