cohere_rerank

Generates vector embeddings to represent input text, using the Cohere API.

Introduced in version 4.37.0.

# Config fields, showing default values
label: ""
cohere_rerank:
  base_url: https://api.cohere.com
  api_key: "" # No default (required)
  model: rerank-v3.5 # No default (required)
  query: "" # No default (required)
  documents: "" # No default (required)
  top_n: "0"
  max_tokens_per_doc: 4096

This processor sends document strings to the Cohere API, which reranks them based on the relevance to the query.

To learn more about reranking, see the Cohere API documentation.

The output of this processor is an array of objects, each containing a “document” field with the original document content, a “relevance_score” field indicating how relevant it is to the query, and an index field that refers to the document’s position within the input documents array. The objects are ordered by their relevance score (highest first).

Examples

Rerank some documents based on a query

Rerank some documents based on a query

input:
  generate:
    interval: 1s
    mapping: |
      root = {
        "query": fake("sentence"),
        "docs": [fake("paragraph"), fake("paragraph"), fake("paragraph")],
      }
pipeline:
  processors:
  - cohere_rerank:
      model: rerank-v3.5
      api_key: "${COHERE_API_KEY}"
      query: "${!this.query}"
      documents: "root = this.docs"
output:
  stdout: {}

Fields

`base_url`

The base URL to use for API requests.

Type: string

Default: "https://api.cohere.com"

`api_key`

The API key for the Cohere API.

Type: string

`model`

The name of the Cohere model to use.

Type: string

# Examples

model: rerank-v3.5

`query`

The search query This field supports interpolation functions.

Type: string

`documents`

A list of texts that will be compared to the query. For optimal performance Cohere recommends against sending more than 1000 documents in a single request. NOTE: structured data should be formatted as YAML for best performance.

Type: string

`top_n`

The number of documents to return, if 0 all documents are returned. This field supports interpolation functions.

Type: string

Default: "0"

`max_tokens_per_doc`

Long documents will be automatically truncated to the specified number of tokens.

Type: int

Default: 4096