Rerank documents

Rerank candidate documents against a query and return relevance scores and original positions.

POST https://token.moi.matrixorigin.cn/v1/rerank

Before you begin

  1. Prepare a personal access token or service account API key for Genesis. See Access credentials for credential selection and permissions.

  2. Use List callable models to select a model that supports Rerank.

Request body

Replace $GENESIS_ACCESS_TOKEN and $MODEL_ID with your credential and the selected model ID.

curl -X POST \
  "https://token.moi.matrixorigin.cn/v1/rerank" \
  -H "Authorization: Bearer $GENESIS_ACCESS_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "'"$MODEL_ID"'",
  "query": "database",
  "documents": [
    "database transaction",
    "weather forecast"
  ],
  "top_n": 1,
  "return_documents": true
}'

The fields below cover text queries and text candidates. Image or video reranking requires corresponding model capabilities and input configuration.

Parameter

Type

Required

Description

model

string

Yes

Selected model ID from data[].id in the model list.

query

string or object

Yes

Nonempty query text used to assess relevance. Object fields are listed below.

query.content

string or array of object

Required for an object query

Text or text blocks for an object input.

query.content[].type

string

Required for a text block

Use text for a text block.

query.content[].text

string

Required for a text block

Nonempty text content.

documents

array of string or array of object

Yes

Nonempty candidates; preserve their original order to identify returned results.

documents[].content

string or array of object

Required for an object document

Text or text blocks for an object input.

documents[].content[].type

string

Required for a text block

Use text for a text block.

documents[].content[].text

string

Required for a text block

Nonempty text content.

top_n

integer

No

Number of highest-ranked results to return; must be positive.

return_documents

boolean

No

Set to true to include candidate content in results.

Successful response

A successful request returns candidate results ordered by relevance.

{
  "id": "rerank-example",
  "model": "MODEL_ID",
  "results": [
    {
      "index": 0,
      "relevance_score": 0.98,
      "document": {
        "text": "database transaction"
      }
    }
  ],
  "usage": {
    "prompt_tokens": 5,
    "completion_tokens": 0,
    "total_tokens": 5
  }
}

Field

Type

Description

id

string

Rerank request identifier.

model

string

Model that processed the request.

results

array of object

Results ordered by relevance.

results[].index

integer

Zero-based position in the original documents.

results[].relevance_score

number

Relevance score assigned by the model.

results[].document

string or object

Included when candidate content is requested; text object fields are listed below.

results[].document.text

string

Candidate text.

usage

object

Token usage returned by the model.

usage.prompt_tokens

integer

Input tokens.

usage.completion_tokens

integer

Generated tokens.

usage.total_tokens

integer

Total input and generated tokens.

Error response

A missing model ID returns HTTP 400. Supply the model ID before sending the request again.

{
  "error": {
    "message": "missing required parameter: model",
    "type": "invalid_request",
    "code": "invalid_request"
  }
}

Field

Type

Description

error

object

Error details.

error.message

string

Error description; the example indicates a missing model ID.

error.type

string

Error type; may be omitted by a model service.

error.code

string or null

Error code; may be omitted by a model service.

Last updated on