Skip to main content

​
Overview

RerankerConfig specifies which reranking strategy to use and how to configure it.

​
Definition

from dataclasses import dataclass, field
from typing import Dict, Any, Optional

@dataclass
class RerankerConfig:
    """Configuration for reranker."""
    type: str = "llm"
    custom_reranker: Optional[BaseReranker] = None
    kwargs: Dict[str, Any] = field(default_factory=dict)

​
Fields

​
type
str
default:"llm"
Reranker type: “llm”, “cohere”, “sentence-transformer”, or “none”
​
custom_reranker
Optional[BaseReranker]
default:"None"
Custom reranker instance (overrides type if provided)
​
kwargs
Dict[str, Any]
default:"{}"
Additional keyword arguments for the reranker

​
Reranker Types

​
LLM-based (Default)

Uses your configured LLM to score and rerank chunks: 
from mini import RerankerConfig

reranker_config = RerankerConfig(
    type="llm"
)
Pros: No additional API required, uses existing LLM
Cons: Uses LLM tokens

​
Cohere

Uses Cohere’s specialized reranking models: 
import os
from mini import RerankerConfig

reranker_config = RerankerConfig(
    type="cohere",
    kwargs={
        "api_key": os.getenv("COHERE_API_KEY"),
        "model": "rerank-english-v3.0"
    }
)
Pros: Highest quality, specialized for reranking
Cons: Requires Cohere API key, additional cost

​
Sentence Transformer

Uses local cross-encoder models: 
from mini import RerankerConfig

reranker_config = RerankerConfig(
    type="sentence-transformer",
    kwargs={
        "model_name": "cross-encoder/ms-marco-MiniLM-L-6-v2",
        "device": "cuda"  # or "cpu"
    }
)
Pros: Local, private, no API calls
Cons: Requires model download, GPU recommended

​
Disable Reranking

from mini import RerankerConfig

reranker_config = RerankerConfig(
    type="none"
)
Pros: Fastest
Cons: Lower quality results

​
Complete Configurations

​
Production (Cohere)

reranker_config = RerankerConfig(
    type="cohere",
    kwargs={
        "api_key": os.getenv("COHERE_API_KEY"),
        "model": "rerank-english-v3.0",
        "max_chunks_per_doc": None
    }
)

​
Local/Private (Sentence Transformer)

reranker_config = RerankerConfig(
    type="sentence-transformer",
    kwargs={
        "model_name": "cross-encoder/ms-marco-MiniLM-L-6-v2",
        "device": "cuda"
    }
)

​
Simple (LLM-based)

reranker_config = RerankerConfig(
    type="llm"
)

​
Custom Reranker

from mini.reranker import CohereReranker

custom_reranker = CohereReranker(
    api_key="your-key",
    model="rerank-multilingual-v3.0"
)

reranker_config = RerankerConfig(
    custom_reranker=custom_reranker
)

​
Usage Example

from mini import AgenticRAG, LLMConfig, RetrievalConfig, RerankerConfig

rag = AgenticRAG(
    vector_store=vector_store,
    embedding_model=embedding_model,
    llm_config=LLMConfig(model="gpt-4o-mini"),
    retrieval_config=RetrievalConfig(
        top_k=10,
        rerank_top_k=3,
        use_reranking=True
    ),
    reranker_config=RerankerConfig(
        type="cohere",
        kwargs={
            "api_key": os.getenv("COHERE_API_KEY"),
            "model": "rerank-english-v3.0"
        }
    )
)

​
Comparison

RerankerQualitySpeedCostPrivacy
Cohere⭐⭐⭐⭐⭐⚡⚡⚡💰💰☁️
LLM⭐⭐⭐⭐⚡⚡💰💰☁️
Sentence Transformer⭐⭐⭐⚡⚡⚡⚡💰🔒
None⭐⭐⚡⚡⚡⚡⚡💰-

​
See Also

Rerankers Overview

Learn about reranking

Cohere Reranker

Cohere API details

Reranking Feature

Reranking guide