Rerank Innovation Release

Suggest edits

This documentation covers the current Innovation Release of EDB Postgres AI. You may also want the docs for the current LTS version.

Overview

Rerank Drivers can be used to rerank search results for a particular query. Every Rerank Driver implements the following methods that can be used directly:

run(query: str, artifacts: list[TextArtifact]) reranks a list of TextArtifact based on the original query.

Rerank Drivers can also be used with a RagEngine's Rerank Module.

Rerank Drivers

Local

The LocalRerankDriver uses a simple relatedness calculation.

Code
Logs

from griptape.artifacts import TextArtifact
from griptape.drivers.rerank.local import LocalRerankDriver

rerank_driver = LocalRerankDriver()

artifacts = rerank_driver.run(
    "What is the capital of France?",
    [
        TextArtifact("Hotdog"),
        TextArtifact("San Francisco"),
        TextArtifact("Paris"),
        TextArtifact("Baguette"),
        TextArtifact("French Hotdog"),
    ],
)
for artifact in artifacts:
    print(artifact.value)

Paris
San Francisco
Baguette
French Hotdog
Hotdog

Cohere

The CohereRerankDriver uses Cohere's Rerank model.

Code
Logs

import os

from griptape.artifacts import TextArtifact
from griptape.drivers.rerank.cohere import CohereRerankDriver

rerank_driver = CohereRerankDriver(
    api_key=os.environ["COHERE_API_KEY"],
)

artifacts = rerank_driver.run(
    "Where is NYC located?",
    [
        TextArtifact("NYC Media"),
        TextArtifact("New York City Police Department"),
        TextArtifact("New York City"),
        TextArtifact("New York City Subway"),
    ],
)
for artifact in artifacts:
    print(artifact.value)

New York City
NYC Media
New York City Police Department
New York City Subway

Nvidia NIM

The NvidiaNimRerankDriver uses the Nvidia NIM Reranking API.

Code

from griptape.artifacts import TextArtifact
from griptape.drivers.rerank.nvidia_nim import NvidiaNimRerankDriver

rerank_driver = NvidiaNimRerankDriver(
    model="nvidia/bert-base-uncased",
    base_url="http://localhost:8000",
)

artifacts = rerank_driver.run(
    "Where is NYC located?",
    [
        TextArtifact("NYC Media"),
        TextArtifact("New York City Police Department"),
        TextArtifact("New York City"),
        TextArtifact("New York City Subway"),
    ],
)
for artifact in artifacts:
    print(artifact.value)

← Prev

Embedding

↑ Up

Prompt

Vector Store