Docs
Embeddings and reranking
Embedding and reranking share one key. Search pipelines often chain the two stages together.
Embedding
Pass a string or an array as input. Arrays are processed in one call and counted as a single request.
| Endpoint | Field | Note |
|---|---|---|
| /embeddings | input | string or string[] |
| /embeddings | dimensions | Optional reduced size |
| /rerank | query | Search query |
| /rerank | documents | Candidate strings |
| /rerank | top_n | Number of results |
Reranking
Pass query and documents to get back indices and scores sorted by relevance.
Dimension reduction
Specify dimensions to receive reduced vectors. It lowers storage cost, but verify retrieval quality.