Vllm serve Reranker

General
1

In the vllmserve service, a Reranker service is deployed. Which of the following interfaces should be used to call it?

Route: /v1/score, Methods: POST

Route: /v1/rerank, Methods: POST

Route: /v2/rerank, Methods: POST

2

To call a reranker service deployed with vllm serve, you should use the /v1/rerank (POST) interface. This endpoint is specifically designed for rerank models and is compatible with Jina AI and Cohere re-rank API interfaces. The /v1/score endpoint is for sentence-pair scoring, not reranking, and /v2/rerank is an alternative version but /v1/rerank is the standard documented route for rerankers in vLLM serving mode.

See official vLLM docs and API server source for confirmation.

Would you like more detail or example usage?

Sources: