Cohere Ranker

The Cohere Ranker leverages Cohere's rerank models to improve result ordering by applying semantic reranking to retrieved candidates.

Unlike retrieval or embedding functions, Cohere Ranker runs as a post-retrieval step. It evaluates the semantic relevance between the query and document text and reorders the candidate results accordingly.

Cohere Ranker is particularly useful when:

• Retrieved results are relevant but not ideally ordered

• Semantic relevance matters more than vector distance alone

• You need multilingual or long-text reranking

Before you start

Before using the Cohere Ranker, make sure the following prerequisites are met.

• Choose a rerank model

  Decide which Cohere rerank model to use, such as rerank-english-v3.0. Your choice determines how semantic relevance is evaluated during reranking. For details, see Cohere official documentation.

• Integrate with Cohere and get your integration ID

  To use Cohere Ranker, you must first integrate Cohere as a model provider in the Zilliz Cloud console. For detailed steps, see Integrate with Model Providers.

• Plan a collection schema with a rerankable text field

  Ensure that your collection includes one VARCHAR field containing the text to be reranked.

Use Cohere Ranker

This section shows how to apply Cohere Ranker during search to rerank retrieved results.

Cohere Ranker is defined and applied at search time, allowing you to enable or disable reranking per query.

Preparations

The following setup prepares a collection and sample data for search and reranking.

Prepare a collection with sample data

from pymilvus import MilvusClient, DataType

client = MilvusClient(
    uri="YOUR_ZILLIZ_CLOUD_URI",
    token="YOUR_ZILLIZ_CLOUD_TOKEN",
)

collection_name = "cohere_rerank_demo"

# Define collection schema 
schema = client.create_schema()

schema.add_field("id", DataType.INT64, is_primary=True, auto_id=False)
schema.add_field("document", DataType.VARCHAR, max_length=1000)
schema.add_field("dense", DataType.FLOAT_VECTOR, dim=4)

# Configure index
index_params = client.prepare_index_params()

index_params.add_index(
    field_name="dense",
    index_type="AUTOINDEX",
    metric_type="COSINE"
)

# Create collection
client.create_collection(
    collection_name=collection_name,
    schema=schema,
    index_params=index_params
)

# Insert sample data
data = [
    {
        "id": 1,
        "document": "Recent renewable energy developments include improved solar efficiency.",
        "dense": [0.10, 0.20, 0.30, 0.40],
    },
    {
        "id": 2,
        "document": "Climate policy and carbon markets have evolved rapidly in recent years.",
        "dense": [0.11, 0.19, 0.28, 0.39],
    },
    {
        "id": 3,
        "document": "New battery technology helps stabilize wind and solar power generation.",
        "dense": [0.90, 0.10, 0.05, 0.02],
    },
    {
        "id": 4,
        "document": "Vector databases support similarity search for machine learning applications.",
        "dense": [0.01, 0.02, 0.03, 0.04],
    },
]

client.insert(collection_name, data)

Define the rerank function

Cohere Ranker is defined at search time, not as part of the collection schema.

The rerank function specifies:

• which text field (VARCHAR) to rerank

• which Cohere rerank model to use

• which query text is evaluated for relevance

from pymilvus import Function, FunctionType

cohere_ranker = Function(
    name="cohere_semantic_ranker",
    input_field_names=["document"],
    function_type=FunctionType.RERANK,
    params={
        "reranker": "model",
        "provider": "cohere",
        "model_name": "rerank-english-v3.0",
        "queries": ["renewable energy developments"],

        "integration_id": "YOUR_INTEGRATION_ID",

    }
)

📘Notes

The number of strings in queries must match the number of queries issued in the search request.

Search with the rerank function

query_vector = [0.12, 0.21, 0.29, 0.41]

results = client.search(
    collection_name=collection_name,
    data=[query_vector],
    anns_field="dense",
    limit=3,
    output_fields=["document"],
    ranker=cohere_ranker,
)

print(results)

During this search:

1. Zilliz Cloud retrieves candidates using vector search.

2. Cohere Ranker evaluates semantic relevance for each candidate.

3. The result set is reordered before being returned.

Next steps

Cohere Ranker can also be used with hybrid search.

Search and hybrid search apply the ranker in the same manner.

In both cases, you pass the rerank function via the ranker parameter at search time.

For details, see Multi-Vector Hybrid Search.