We’ll use Cohere’s Python SDKs (cohere and cohere-aws) for the code examples. Follow along in this notebook.

Introduction

Reranking is an essential technique in information retrieval systems, especially in large-scale search applications. It is a process of reordering a set of initially retrieved documents based on their relevance to a user's query.

With the vast amount of information available, reranking helps refine and improve the quality of search results, ensuring that the most relevant and useful documents are presented to the user. Cohere’s reranking models offer a powerful solution to enhance the search experience.

What’s great about reranking is that while it gives a huge boost to search results, implementing Cohere’s Rerank models requires adding only one line of code to any existing search system, whether it’s a semantic search system or a traditional search system that uses keyword-based approaches.

For example, a major provider of digital workforce collaboration and productivity tools found that their existing search tools did not perform well across customer-generated knowledge bases, and the large volume of new customer-generated content meant embedding was not viable.

Cohere Rerank was integrated with the customer’s search systems. Rerank used output from existing search tools and reordered the results for use by the company’s Q&A app to provide relevant and more accurate answers.

Implementing a reranking system at a provider of digital workforce collaboration and productivity tools

How Cohere Rerank Works

The Cohere Rerank endpoint takes a query and a response and outputs a relevance score between them. This enables the retrieval of the most relevant documents to a particular query. For example, when searching for "Regulatory approval", the initial results may include documents that only tangentially mention the topic. However, by applying Rerank, the top results are those with the highest relevance scores, directly answering the query.

Rerank improves the ordering of documents based on relevance to a query

Rerank 3 is Cohere’s latest reranking model, which provides state-of-the-art capabilities for enterprise search, including:

4k context length to significantly improve search quality for longer documents
Ability to search over multi-aspect and semi-structured data, such as emails, invoices, JSON documents, code, and tables
Multilingual coverage of 100+ languages
Improved latency and lower total cost of ownership (TCO)

In this chapter, we'll explore how to use the Cohere Rerank endpoint on Amazon SageMaker. To see the most updated list of available models on SageMaker, visit Cohere’s listing on the AWS Marketplace.

Code Walkthrough

Let’s look at an example of a multi-aspect search on semi-structured data, and walk through how to perform reranking on email data that contains multiple fields: “title” and “content.”

Setup

To get set up with SageMaker, follow the steps outlined in the Cohere documentation, which takes you through the prerequisites and getting access to Cohere models.

Let’s now install and import the necessary libraries and set up our Cohere client.

We’ll need to create a SageMaker endpoint that exposes access to a Cohere model (Rerank v3 in our case). For this, we’ll use the cohere_aws SDK which makes it easy to set up the endpoint, together with AWS’s boto3 library.

# ! pip install cohere cohere-aws boto3

import os
import boto3
import cohere
import cohere_aws
from cohere_aws import Client

Once the endpoint is created (as we’ll walk through later), we can access it using the Cohere SDK. To do this, let’s create a SagemakerClient by passing the necessary AWS credentials.

import cohere

# Create SageMaker client via the native Cohere SDK
# Contact your AWS administrator for the credentials
co = cohere.SagemakerClient(
    aws_region="YOUR_AWS_REGION",
    aws_access_key="YOUR_AWS_ACCESS_KEY_ID",
    aws_secret_key="YOUR_AWS_SECRET_ACCESS_KEY",
    aws_session_token="YOUR_AWS_SESSION_TOKEN",
)

# For creating an endpoint, you need to use the cohere_aws client: Set environment variables with the AWS credentials
os.environ['AWS_ACCESS_KEY_ID'] = "YOUR_AWS_ACCESS_KEY_ID"
os.environ['AWS_SECRET_ACCESS_KEY'] = "YOUR_AWS_SECRET_ACCESS_KEY"
os.environ['AWS_SESSION_TOKEN'] = "YOUR_AWS_SESSION_TOKEN"

The next step is to define the model package Amazon Resource Names (ARN) for the Rerank model. The ARN is an identifying string for a SageMaker resource, and it varies between the regions where a resource is available.

Here, we define the Cohere package for the Rerank model and map the model package against each region, which gives the complete ARN for each region.

# Create SageMaker endpoint via the cohere_aws SDK
cohere_package = "cohere-rerank-english-v3-01-d3687e0d2e3a366bb904275616424807"
model_package_map = {
    "us-east-1": f"arn:aws:sagemaker:us-east-1:865070037744:model-package/{cohere_package}",
    "us-east-2": f"arn:aws:sagemaker:us-east-2:057799348421:model-package/{cohere_package}",
    "us-west-1": f"arn:aws:sagemaker:us-west-1:382657785993:model-package/{cohere_package}",
    "us-west-2": f"arn:aws:sagemaker:us-west-2:594846645681:model-package/{cohere_package}",
    "ca-central-1": f"arn:aws:sagemaker:ca-central-1:470592106596:model-package/{cohere_package}",
    "eu-central-1": f"arn:aws:sagemaker:eu-central-1:446921602837:model-package/{cohere_package}",
    "eu-west-1": f"arn:aws:sagemaker:eu-west-1:985815980388:model-package/{cohere_package}",
    "eu-west-2": f"arn:aws:sagemaker:eu-west-2:856760150666:model-package/{cohere_package}",
    "eu-west-3": f"arn:aws:sagemaker:eu-west-3:843114510376:model-package/{cohere_package}",
    "eu-north-1": f"arn:aws:sagemaker:eu-north-1:136758871317:model-package/{cohere_package}",
    "ap-southeast-1": f"arn:aws:sagemaker:ap-southeast-1:192199979996:model-package/{cohere_package}",
    "ap-southeast-2": f"arn:aws:sagemaker:ap-southeast-2:666831318237:model-package/{cohere_package}",
    "ap-northeast-2": f"arn:aws:sagemaker:ap-northeast-2:745090734665:model-package/{cohere_package}",
    "ap-northeast-1": f"arn:aws:sagemaker:ap-northeast-1:977537786026:model-package/{cohere_package}",
    "ap-south-1": f"arn:aws:sagemaker:ap-south-1:077584701553:model-package/{cohere_package}",
    "sa-east-1": f"arn:aws:sagemaker:sa-east-1:270155090741:model-package/{cohere_package}",
}

region = boto3.Session().region_name

if region not in model_package_map.keys():
    raise Exception("UNSUPPORTED REGION")

model_package_arn = model_package_map[region]

The ARN for each model is available in the Marketplace once you subscribe to the model and proceed to configure it. The screenshot below shows the Rerank 3 English listing as an example.

Rerank 3 English model listing on the AWS Marketplace

Create Endpoint

With SageMaker, we’ll need to create an endpoint via an AWS instance. The marketplace listing provides more details, including pricing, on the recommended instance type for a particular model.

To create the endpoint, we define:

arn: The model package ARN we defined in the previous step
endpoint_name: A name we can give as an identifier
instance_type: The instance type to be used
n_instances: The number of instances

We pass the arguments to the create_endpoint method from the cohere_aws library.

co_aws = Client(region_name=region)

co_aws.create_endpoint(arn=model_package_arn, endpoint_name="my-rerank-v3", instance_type="ml.g5.xlarge", n_instances=1)

The following output indicates that the process of creating an endpoint is complete.

---------!

Retrieve Documents

Let’s assume that the first stage of retrieval has already been performed, whether it’s through a semantic, keyword, or any other type of search system.

Here we have a list of nine documents that represent the search results of that first stage. Each document has two fields, Title and Content, corresponding to the contents of an email. Each email is a dictionary containing these fields that preserve its semi-structured format, which the Rerank endpoint can use.

documents = [
    {"Title":"Incorrect Password","Content":"Hello, I have been trying to access my account for the past hour and it keeps saying my password is incorrect. Can you please help me?"},
    {"Title":"Confirmation Email Missed","Content":"Hi, I recently purchased a product from your website but I never received a confirmation email. Can you please look into this for me?"},
    {"Title":"Questions about Return Policy","Content":"Hello, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective."},
    {"Title":"Customer Support is Busy","Content":"Good morning, I have been trying to reach your customer support team for the past week but I keep getting a busy signal. Can you please help me?"},
    {"Title":"Received Wrong Item","Content":"Hi, I have a question about my recent order. I received the wrong item and I need to return it."},
    {"Title":"Customer Service is Unavailable","Content":"Hello, I have been trying to reach your customer support team for the past hour but I keep getting a busy signal. Can you please help me?"},
    {"Title":"Return Policy for Defective Product","Content":"Hi, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective."},
    {"Title":"Wrong Item Received","Content":"Good morning, I have a question about my recent order. I received the wrong item and I need to return it."},
    {"Title":"Return Defective Product","Content":"Hello, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective."}
]

Rerank Documents

To use the endpoint, we now use the cohere SDK. Adding a reranking component is simple with Cohere Rerank. It takes just one line of code to implement.

Calling the Rerank endpoint requires the following arguments:

documents: The list of documents, which we defined in the previous section
query: The user query; we’ll use 'What emails have been about refunds?' as an example
rank_fields: The list of fields that the data contains, which in our case are Title and Content
top_n: The number of documents we want to be returned, sorted from the most to the least relevant document
model: The endpoint_name we defined earlier, which in our case is my-rerank-v3

query = 'What emails have been about refunds?'

response = co.rerank(documents=documents,
                     query=query,
                     rank_fields=["Title","Content"],
                     top_n=3,
                     model="my-rerank-v3")

View Results

Since we defined top_n=3, we’ll get the top three most relevant documents to the query. For each document, the response contains the index of its position in the original list and its relevance score against the query.

Let’s print the results and see the outcome.

print("Documents","\n")

for idx,doc in enumerate(response.results):
    print(f"#{idx+1}:\n{documents[doc.index]}\n")

The search query was looking for emails about refunds. But none of the documents mention the word “refunds” specifically.

However, the Rerank model was able to retrieve the right documents. Some of the documents mentioned the word “return.” The Rerank model can capture semantically similar meanings between two pieces of text, so it can return documents that mention return instead, which has a very similar meaning to return.

Documents 

#1:
{'Title': 'Questions about Return Policy', 'Content': 'Hello, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective.'}

#2:
{'Title': 'Return Policy for Defective Product', 'Content': 'Hi, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective.'}

#3:
{'Title': 'Return Defective Product', 'Content': 'Hello, I have a question about the return policy for this product. I purchased it a few weeks ago and it is defective.'}

Important Note: You will continue to incur charges as long as an endpoint is running, so remember to delete it when your usage ends.

co_aws.delete_endpoint()
co_aws.close()

Conclusion

Reranking is a valuable technique used in information retrieval systems to enhance the relevance of search results. Cohere's Rerank endpoint, including its latest model, Rerank 3, offers improved capabilities for enterprise search.

By incorporating reranking with a single line of code, as shown in our example in this chapter, the model successfully identified semantically similar documents, even when specific keywords were absent from the query. This example highlights the potential benefits of integrating reranking into existing search systems to enhance search accuracy and user satisfaction.

In Chapter 6, we’ll explore how to build a retrieval-augmented generation (RAG) application using Chat and Embed on Amazon Bedrock and Rerank on Amazon SageMaker.

Contents

Reranking Using Cohere Rerank on Amazon SageMaker