Implementation:Langchain ai Langchain Chroma Max Marginal Relevance Search

Knowledge Sources	LangChain
Domains	Information_Retrieval, Vector_Search
Last Updated	2026-02-11 00:00 GMT

Overview

Concrete tool for performing MMR search against a Chroma vector store provided by the LangChain Chroma integration.

Description

The Chroma.max_marginal_relevance_search() method first fetches fetch_k candidate documents by similarity, then applies the MMR algorithm to select k results that balance relevance with diversity. The lambda_mult parameter (0.0 to 1.0) controls this balance.

Usage

Use instead of similarity_search() when you need diverse results, particularly for RAG contexts.

Code Reference

Source Location

• Repository: langchain
• File: libs/partners/chroma/langchain_chroma/vectorstores.py
• Lines: L1075-1122

Signature

def max_marginal_relevance_search(
    self,
    query: str,
    k: int = DEFAULT_K,
    fetch_k: int = 20,
    lambda_mult: float = 0.5,
    filter: dict[str, str] | None = None,
    where_document: dict[str, str] | None = None,
    **kwargs: Any,
) -> list[Document]:

Import

from langchain_chroma import Chroma

I/O Contract

Inputs

Name	Type	Required	Description
query	str	Yes	Search query
k	int	No (default: 4)	Final results count
fetch_k	int	No (default: 20)	Initial candidates to fetch
lambda_mult	float	No (default: 0.5)	Diversity/relevance balance (0=diversity, 1=relevance)
filter	dict or None	No	Metadata filter

Outputs

Name	Type	Description
return	list[Document]	Top-k documents selected via MMR

Usage Examples

MMR Search

# Diverse results (more diversity)
results = vectorstore.max_marginal_relevance_search(
    "machine learning applications",
    k=5,
    fetch_k=20,
    lambda_mult=0.3,  # Favor diversity
)

# Relevant results (more relevance)
results = vectorstore.max_marginal_relevance_search(
    "machine learning applications",
    k=5,
    fetch_k=20,
    lambda_mult=0.8,  # Favor relevance
)

Related Pages

Implements Principle

• Principle:Langchain_ai_Langchain_Maximal_Marginal_Relevance