Rerank

A rerank model scores a list of documents against a query and returns them ordered by relevance. It is the second stage of a retrieval pipeline: a vector search brings back a few dozen candidates, the reranker reads each one against the question and keeps the best. The endpoint is available over the REST API. The documents to score travel as input:

curl https://api.gwenlake.com/v1/rerank \
  -H "Authorization: Bearer $GWENLAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bge-reranker-v2-m3",
    "query": "What is the refund policy?",
    "input": ["Shipping takes 3 days.", "Refunds are accepted within 30 days."],
    "top_n": 1
  }'

Field	Meaning
`model`	a rerank model id, as `GET /models` lists them
`query`	the question
`input`	the documents to score, as strings
`top_n`	keep only the best `top_n`; omit to score them all

The answer lists the documents by decreasing relevance, each with its position in input and its score, so the caller maps the result back to its own records.

With the Python library

The library has no rerank resource yet; call the endpoint with the client’s transport, which carries your credentials:

from gwenlake import Gwenlake
from gwenlake.client import RequestOptions

client = Gwenlake()
response = client._client.send(RequestOptions(
    method="POST",
    url="/rerank",
    json_data={"model": "bge-reranker-v2-m3", "query": "What is the refund policy?",
               "input": ["Shipping takes 3 days.", "Refunds are accepted within 30 days."], "top_n": 1},
))
response.raise_for_status()
print(response.json())

Choosing a model

client.models.list() returns every inference model you can call, rerank models included. Model ids are stable; the catalog of available models evolves with the platform.

Get started

Concepts

Workspace

Data

Code

Build

Models

Inference

Agents

Client

Changelog

With the Python library

Choosing a model

​With the Python library

​Choosing a model

With the Python library

Choosing a model