Implementation:Marker Inc Korea AutoRAG CohereReranker

Knowledge Sources	Marker_Inc_Korea_AutoRAG
Domains	RAG, Passage_Reranking, API
Last Updated	2026-02-12 12:00 GMT

Overview

CohereReranker reranks passages using Cohere's cloud-based rerank API endpoint with asynchronous batch processing.

Description

The CohereReranker class extends BasePassageReranker and leverages the Cohere rerank API to score and reorder retrieved passages based on their relevance to a given query. It initializes an asynchronous Cohere client (cohere.AsyncClientV2) using an API key sourced from the constructor parameter or the COHERE_API_KEY / CO_API_KEY environment variables. The _pure method dispatches concurrent reranking tasks per query using the process_batch utility, which respects the configurable batch size to control API throughput.

Usage

Import and use CohereReranker when you need a cloud-based, API-driven passage reranking step in an AutoRAG pipeline. It is suitable when you prefer not to run local GPU inference and have a valid Cohere API key. The default model is rerank-v3.5, with multilingual and English-specific v3.0 variants also available.

Code Reference

Source Location

• Repository: Marker_Inc_Korea_AutoRAG
• File: autorag/nodes/passagereranker/cohere.py
• Lines: 12-118

Signature

class CohereReranker(BasePassageReranker):
    def __init__(self, project_dir: str, *args, **kwargs)
    def _pure(
        self,
        queries: List[str],
        contents_list: List[List[str]],
        scores_list: List[List[float]],
        ids_list: List[List[str]],
        top_k: int,
        batch: int = 64,
        model: str = "rerank-v3.5",
    ) -> Tuple[List[List[str]], List[List[str]], List[List[float]]]

Import

from autorag.nodes.passagereranker.cohere import CohereReranker

I/O Contract

Inputs

Name	Type	Required	Description
queries	List[str]	Yes	List of query strings to use for reranking.
contents_list	List[List[str]]	Yes	List of passage content lists corresponding to each query.
scores_list	List[List[float]]	Yes	List of initial retrieval score lists (not used in scoring, passed through for interface compatibility).
ids_list	List[List[str]]	Yes	List of passage ID lists corresponding to each query.
top_k	int	Yes	Number of top passages to return after reranking.
batch	int	No	Number of concurrent API requests to process in a batch. Default is 64.
model	str	No	Cohere rerank model name. Options include "rerank-v3.5", "rerank-english-v3.0", "rerank-multilingual-v3.0". Default is "rerank-v3.5".

Outputs

Name	Type	Description
content_result	List[List[str]]	Reranked passage contents per query, ordered by relevance.
id_result	List[List[str]]	Passage IDs corresponding to the reranked contents.
score_result	List[List[float]]	Relevance scores assigned by the Cohere rerank model.

Usage Examples

Basic Usage

import os
os.environ["COHERE_API_KEY"] = "your-cohere-api-key"

from autorag.nodes.passagereranker.cohere import CohereReranker

reranker = CohereReranker(project_dir="/path/to/project")

queries = ["What is retrieval augmented generation?"]
contents_list = [["RAG combines retrieval and generation.", "Unrelated passage.", "RAG improves accuracy."]]
scores_list = [[0.9, 0.5, 0.8]]
ids_list = [["id1", "id2", "id3"]]

contents, ids, scores = reranker._pure(
    queries=queries,
    contents_list=contents_list,
    scores_list=scores_list,
    ids_list=ids_list,
    top_k=2,
    model="rerank-v3.5",
)
print(contents)  # Top 2 passages reranked by Cohere
print(scores)    # Corresponding relevance scores

Related Pages

• Environment:Marker_Inc_Korea_AutoRAG_API_Keys_Configuration