What causes "rerank returned %d indices for %d documents" in siyuan-note/siyuan?

Confirm the provider implements the standard {results:[{index,relevance_score}]} schema with 0-based indices.

rerank returned %d indices for %d documents

Error message

rerank returned %d indices for %d documents

What it means

TestRerankModel sends two trivial documents and expects exactly two indices back. If the count differs, the provider is applying a default top_n (returning fewer), returning scores without indices, or returning out-of-range indices that Rerank's loop filters out. The probe treats such a provider as non-conformant for the connectivity test.

Source

Thrown at kernel/util/openai.go:605

func truncateRerankDocument(document string) string {
	runes := []rune(document)
	if len(runes) > rerankDocTextMaxRunes {
		return string(runes[:rerankDocTextMaxRunes])
	}
	return document
}

// TestRerankModel 测试重排模型可用性，用极简 query+documents 发一次重排请求验证连通性与鉴权。
// 返回值：matched 表示是否连通成功，err 为请求错误（鉴权失败、网络异常、模型不存在等，原样返回便于调用方展示原因）。
func TestRerankModel(apiKey, apiBaseURL, model string, timeout int) (matched bool, err error) {
	documents := []string{"a", "b"}
	indices, _, err := Rerank("1", documents, apiKey, apiBaseURL, model, len(documents), timeout)
	if nil != err {
		return
	}
	if len(indices) != len(documents) {
		err = fmt.Errorf("rerank returned %d indices for %d documents", len(indices), len(documents))
		return
	}
	seen := make(map[int]bool, len(indices))
	for _, index := range indices {
		if seen[index] {
			err = fmt.Errorf("rerank returned duplicate index %d", index)
			return
		}
		seen[index] = true
	}
	matched = true
	return
}

// PrepareModelImage 校验并按需缩放图片，尽量保留多模态模型支持的原始格式和图片质量。
func PrepareModelImage(data []byte, maxBytes, maxPixels, maxEdge int) (PreparedImage, error) {
	if len(data) == 0 {
		return PreparedImage{}, errors.New("image data is empty")

View on GitHub (pinned to 251596fc0d)

Solutions

Confirm the provider implements the standard {results:[{index,relevance_score}]} schema with 0-based indices.
Inspect the raw rerank response to see whether indices are 0-based and complete.
Verify the model name actually supports rerank.
If non-conformant, use a different provider or configure rerank manually without relying on the probe.

Defensive patterns

Strategy: try-catch

Try / catch

if ok, err := util.TestRerankModel(key, endpoint, model, 30); err != nil || !ok {
    log.Warnf("rerank provider may be non-conformant: %s; falling back to no rerank", err)
}

Prevention

Prefer providers that return all documents with 0-based indices
Inspect the rerank response schema before adopting a provider
Treat rerank as optional and degrade to original ranking on failure

When it happens

Trigger: Rerank provider that caps results at a default top_n=1; provider returning scores without an index field; provider returning 1-based indices (filtered out as out-of-range for a 2-doc input).

Common situations: Provider whose default top_n is 1 and ignores the top_n=len(docs) hint; non-standard response schema; provider that omits low-relevance documents.

Related errors

AI-assisted analysis of siyuan-note/siyuan@251596fc0d (2026-08-12). Data as JSON: /api/errors/f28ad574b03a93db. Report an issue: GitHub.