Skip to main content
Open In ColabOpen on GitHub

Pinecone Rerank

This notebook shows how to use PineconeRerank for two-stage vector retrieval reranking using Pinecone's hosted reranking API as demonstrated in langchain_pinecone/libs/pinecone/rerank.py.

Setup​

Install the langchain-pinecone package.

%pip install -qU "langchain-pinecone"

Credentials​

Set your Pinecone API key to use the reranking API.

import os
from getpass import getpass

os.environ["PINECONE_API_KEY"] = os.getenv("PINECONE_API_KEY") or getpass(
    "Enter your Pinecone API key: "
)

Instantiation​

Use PineconeRerank to rerank a list of documents by relevance to a query.

from langchain_core.documents import Document
from langchain_pinecone import PineconeRerank

# Initialize reranker
reranker = PineconeRerank(model="bge-reranker-v2-m3")

# Sample documents
documents = [
    Document(page_content="Paris is the capital of France."),
    Document(page_content="Berlin is the capital of Germany."),
    Document(page_content="The Eiffel Tower is in Paris."),
]

# Rerank documents
query = "What is the capital of France?"
reranked_docs = reranker.compress_documents(documents, query)

# Print results
for doc in reranked_docs:
    score = doc.metadata.get("relevance_score")
    print(f"Score: {score:.4f} | Content: {doc.page_content}")
API Reference:Document | PineconeRerank
/Users/jakit/customers/aurelio/langchain-pinecone/libs/pinecone/.venv/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm
``````output
Score: 0.9998 | Content: Paris is the capital of France.
Score: 0.1950 | Content: The Eiffel Tower is in Paris.
Score: 0.0042 | Content: Berlin is the capital of Germany.

Usage​

Reranking with Top-N​

Specify top_n to limit the number of returned documents.

# Return only top-1 result
reranker_top1 = PineconeRerank(model="bge-reranker-v2-m3", top_n=1)
top1_docs = reranker_top1.compress_documents(documents, query)
print("Top-1 Result:")
for doc in top1_docs:
    print(f"Score: {doc.metadata['relevance_score']:.4f} | Content: {doc.page_content}")
Top-1 Result:
Score: 0.9998 | Content: Paris is the capital of France.

Reranking with Custom Rank Fields​

If your documents are dictionaries or have custom fields, use rank_fields to specify the field to rank on.

# Sample dictionary documents with 'text' field
docs_dict = [
    {
        "id": "doc1",
        "text": "Article about renewable energy.",
        "title": "Renewable Energy",
    },
    {"id": "doc2", "text": "Report on economic growth.", "title": "Economic Growth"},
    {
        "id": "doc3",
        "text": "News on climate policy changes.",
        "title": "Climate Policy",
    },
]

# Initialize reranker with rank_fields
reranker_text = PineconeRerank(model="bge-reranker-v2-m3", rank_fields=["text"])
climate_docs = reranker_text.rerank(docs_dict, "Latest news on climate change.")

# Show IDs and scores
for res in climate_docs:
    print(f"ID: {res['id']} | Score: {res['score']:.4f}")
ID: doc3 | Score: 0.9892
ID: doc1 | Score: 0.0006
ID: doc2 | Score: 0.0000

We can rerank based on title field

economic_docs = reranker_text.rerank(docs_dict, "Economic forecast.")

# Show IDs and scores
for res in economic_docs:
    print(
        f"ID: {res['id']} | Score: {res['score']:.4f} | Title: {res['document']['title']}"
    )
ID: doc2 | Score: 0.8918 | Title: Economic Growth
ID: doc3 | Score: 0.0002 | Title: Climate Policy
ID: doc1 | Score: 0.0000 | Title: Renewable Energy

Reranking with Additional Parameters​

You can pass model-specific parameters (e.g., truncate) directly to .rerank().

How to handle inputs longer than those supported by the model. Accepted values: END or NONE. END truncates the input sequence at the input token limit. NONE returns an error when the input exceeds the input token limit.

# Rerank with custom truncate parameter
docs_simple = [
    {"id": "docA", "text": "Quantum entanglement is a physical phenomenon..."},
    {"id": "docB", "text": "Classical mechanics describes motion..."},
]

reranked = reranker.rerank(
    documents=docs_simple,
    query="Explain the concept of quantum entanglement.",
    truncate="END",
)
# Print reranked IDs and scores
for res in reranked:
    print(f"ID: {res['id']} | Score: {res['score']:.4f}")
ID: docA | Score: 0.6950
ID: docB | Score: 0.0001

Use within a chain​

API reference​

  • PineconeRerank(model, top_n, rank_fields, return_documents)
  • .rerank(documents, query, rank_fields=None, model=None, top_n=None, truncate="END")
  • .compress_documents(documents, query) (returns Document objects with relevance_score in metadata)